def compare_histogram(a_hist, b_hist, method=cv2.HISTCMP_CHISQR): """ Compares two histograms, whether they represent greyscale images or colored ones. :param a_hist: image A histogram. :param b_hist: image B histogram. :param method: Comparison method. May be one of the following: CV_COMP_CORREL - Correlation CV_COMP_CHISQR - Chi-Square CV_COMP_INTERSECT - Intersection CV_COMP_BHATTACHARYYA - Bhattacharyya distance CV_COMP_HELLINGER - Synonym for CV_COMP_BHATTACHARYYA Please refer to OpenCV documentation for further details. :return: The result of the comparison between two histograms. """ if isinstance(a_hist, list) and isinstance(b_hist, list): diff = [] for i, channel in enumerate(CHANNELS): diff += [cv2.compareHist(a_hist[i], b_hist[i], method=method)] else: diff = cv2.compareHist(a_hist, b_hist, method=cv2.HISTCMP_CHISQR) return np.mean(diff)
def hist_similarity(image_1, image_2):
"""color hist based image similarity
@param image_1: np.array(the first input image)
@param image_2: np.array(the second input image)
@return similarity: float(range from [0,1], the bigger the more similar)
"""
if image_1.ndim == 2 and image_2.ndim == 2:
hist_1 = cv2.calcHist([image_1], [0], None, [256], [0.0, 255.0])
hist_2 = cv2.calcHist([image_2], [0], None, [256], [0.0, 255.0])
similarity = cv2.compareHist(hist_1, hist_2, cv2.cv.CV_COMP_CORREL)
elif image_1.ndim == 3 and image_2.ndim == 3:
"""R,G,B split"""
b_1, g_1, r_1 = cv2.split(image_1)
b_2, g_2, r_2 = cv2.split(image_2)
hist_b_1 = cv2.calcHist([b_1], [0], None, [256], [0.0, 255.0])
hist_g_1 = cv2.calcHist([g_1], [0], None, [256], [0.0, 255.0])
hist_r_1 = cv2.calcHist([r_1], [0], None, [256], [0.0, 255.0])
hist_b_2 = cv2.calcHist([b_2], [0], None, [256], [0.0, 255.0])
hist_g_2 = cv2.calcHist([g_2], [0], None, [256], [0.0, 255.0])
hist_r_2 = cv2.calcHist([r_2], [0], None, [256], [0.0, 255.0])
similarity_b = cv2.compareHist(hist_b_1,hist_b_2,cv2.cv.CV_COMP_CORREL)
similarity_g = cv2.compareHist(hist_g_1,hist_g_2,cv2.cv.CV_COMP_CORREL)
similarity_r = cv2.compareHist(hist_r_1,hist_r_2,cv2.cv.CV_COMP_CORREL)
sum_bgr = similarity_b + similarity_g + similarity_r
similarity = sum_bgr/3.
else:
gray_1 = cv2.cvtColor(image_1,cv2.cv.CV_RGB2GRAY)
gray_2 = cv2.cvtColor(image_2,cv2.cv.CV_RGB2GRAY)
hist_1 = cv2.calcHist([gray_1], [0], None, [256], [0.0, 255.0])
hist_2 = cv2.calcHist([gray_2], [0], None, [256], [0.0, 255.0])
similarity = cv2.compareHist(hist_1, hist_2, cv2.cv.CV_COMP_CORREL)
return similarity
def compute_surrounding_color_contrast(self, img, rect, container_rect, layers=2):
"""
Measure of the dissimilarity of a window to its immediate surrounding area based on chi-squared
distances of LAB space histogram. See "Measuring the objectness of image windows" Alexe et al.
:type img: ndarray
:type rect: tuple
:type container_rect: tuple
:type layers: int
:rtype : ndarray
:param img: Source image. This is not a cropped ROI image.
:param rect: ROI (x, y, w, h) to compute the color contrast against surroundings.
:param container_rect: Container border rectangle (x, y, w, h) pixels outside this will be ignored.
:param layers: number of surrounding layers to compute contrasts.
:return: Color contrast feature vector of shape (layers+1, ). The last cell contains the contrast
between theROI and the entire image.
"""
# chi-squared distances between window image and its surroundings
cc_feature_vector = np.zeros(layers + 1)
rect_img = crop_image(img, rect)
window_hist = compute_lab_histogram(rect_img)
prev_rect = rect
for level in range(layers):
expanded_rect = expand_rect(prev_rect, radius_increase=1.5)
# crop parts outside the container rect
expanded_rect = get_intersecting_rect2(expanded_rect, container_rect)
expanded_rect_img = crop_image(img, expanded_rect)
# mask of only the immediate surrounding area
ew, eh = expanded_rect_img.shape[1], expanded_rect_img.shape[0]
mask = get_mask((ew, eh), expanded_rect, prev_rect)
# LAB histogram of the surrounding area
surrounding_hist = compute_lab_histogram(expanded_rect_img, mask=mask)
# method=1 is the same as cv.CV_COMP_CHISQR for computing the chi-squared distance
cc_feature_vector[level] = cv2.compareHist(window_hist, surrounding_hist, method=1)
prev_rect = expanded_rect
# cache the background histogram to save computation time
img_id = id(img)
if img_id not in self.lab_hist_cache:
rect_img = crop_image(img, container_rect)
background_hist = compute_lab_histogram(rect_img)
self.lab_hist_cache[img_id] = background_hist
else:
background_hist = self.lab_hist_cache[img_id]
# color contract between the window image and the entire source image
background_hist_dist = cv2.compareHist(window_hist, background_hist, method=1)
cc_feature_vector[layers] = background_hist_dist
return cc_feature_vector
def calcColorHistogramSimilarity(img1, img2):
h_b1, h_g1, h_r1 = calcRGBHistogram(img1)
h_b2, h_g2, h_r2 = calcRGBHistogram(img2)
bSimilarity = cv2.compareHist(h_b1, h_b2, cv2.cv.CV_COMP_CORREL)
gSimilarity = cv2.compareHist(h_g1, h_g2, cv2.cv.CV_COMP_CORREL)
rSimilarity = cv2.compareHist(h_r1, h_r2, cv2.cv.CV_COMP_CORREL)
return (bSimilarity + gSimilarity + rSimilarity)/3.0
def get_distance(self, query_hist, ref_hist, w=[1.0, 1.0, 1.0]):
result = dict()
r0, v0 = query_hist.features.items()[0]
for r1, v1 in ref_hist.features.items():
result[r1] = cv2.compareHist(v0[0], v1[0], cv2.cv.CV_COMP_CORREL) * w[0] + \
cv2.compareHist(v0[1], v1[1], cv2.cv.CV_COMP_CORREL) * w[1] + \
cv2.compareHist(v0[2], v1[2], cv2.cv.CV_COMP_CORREL) * w[2]
return result
def processImg(self,img):
print("processing")
hsv = cv2.cvtColor(img, cv2.COLOR_RGB2HSV)
contourList = []
maskG = cv2.inRange(hsv, np.array([35,100,100]), np.array([70, 255, 255]))
maskG = self.blur(maskG)
maskR = cv2.inRange(hsv, np.array([0,100,100]), np.array([15, 255, 255]))
maskR = self.blur(maskR)
maskB = cv2.inRange(hsv, np.array([90,150,150]), np.array([130, 255, 255]))
maskB = self.blur(maskB)
maskY = cv2.inRange(hsv, np.array([23,100,160]), np.array([30, 255, 255]))
maskY = self.blur(maskY)
maskP1 = cv2.inRange(hsv, np.array([0,50,230]), np.array([10, 150, 255]))
maskP2 = cv2.inRange(hsv, np.array([150,50,230]), np.array([180,150,255]))
maskP = maskP1+maskP2
maskP = self.blur(maskP)
contoursG = cv2.findContours(maskG, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contoursR = cv2.findContours(maskR, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contoursB = cv2.findContours(maskB, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contoursY = cv2.findContours(maskY, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contoursP = cv2.findContours(maskP, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
self.contourAppend(contourList,contoursG,"green")
self.contourAppend(contourList,contoursR,"red")
self.contourAppend(contourList,contoursB,"blue")
self.contourAppend(contourList,contoursY,"yellow")
self.contourAppend(contourList,contoursP,"pink")
if len(contourList)>0:
biggest = self.findBiggest(contourList)
else:
biggest = None
if biggest != None:
print(biggest.text)
#print(biggest.contour)
cv2.drawContours(img, biggest.contour, -1, (0, 255, 0), 3)
#cv2.imshow("oooo",img)
#cv2.waitKey(0)
if biggest.text != "pink":
self.saveImg(img,biggest.text)
elif biggest.text == "lol":
x,y,w,h = cv2.boundingRect(biggest.contour)
sliced = hsv[x:x+w,y:y+h,:]
hsvTest = cv2.calcHist(sliced,[0,1],None,[180,256],ranges)
racecarVal = cv2.compareHist(hsvTest,self.racecar,cv2.cv.CV_COMP_CORREL)
ariVal = cv2.compareHist(hsvTest,self.ari,cv2.cv.CV_COMP_CORREL)
sertacVal = cv2.compareHist(hsvTest,self.sertac,cv2.cv.CV_COMP_CORREL)
catVal = cv2.compareHist(hsvTest,self.cat,cv2.cv.CV_COMP_CORREL)
maxVal = max(racecarVal,ariVal,sertacVal,catVal)
if maxVal == racecarVal:
self.saveImg(img,"racecar")
elif maxVal == ariVal:
self.saveImg(img,"ari")
elif maxVal == sertacVal:
self.saveImg(img,"sertac")
else:
self.saveImg(img,"cat")
def returnHistogramComparison(self, hist_1, hist_2, method='intersection'):
"""Return the comparison value of two histograms.
Comparing an histogram with itself return 1.
@param hist_1
@param hist_2
@param method the comparison method.
intersection: (default) the histogram intersection (Swain, Ballard)
"""
if cv2.__version__.split(".")[0] == '3':
if(method=="intersection"):
comparison = cv2.compareHist(hist_1, hist_2, cv2.HISTCMP_INTERSECT)
elif(method=="correlation"):
comparison = cv2.compareHist(hist_1, hist_2, cv2.HISTCMP_CORREL)
elif(method=="chisqr"):
comparison = cv2.compareHist(hist_1, hist_2, cv2.HISTCMP_CHISQR)
elif(method=="bhattacharyya"):
comparison = cv2.compareHist(hist_1, hist_2, cv2.HISTCMP_BHATTACHARYYA)
else:
raise ValueError('[DEEPGAZE] color_classification.py: the method specified ' + str(method) + ' is not supported.')
else:
if(method=="intersection"):
comparison = cv2.compareHist(hist_1, hist_2, cv2.cv.CV_COMP_INTERSECT)
elif(method=="correlation"):
comparison = cv2.compareHist(hist_1, hist_2, cv2.cv.CV_COMP_CORREL)
elif(method=="chisqr"):
comparison = cv2.compareHist(hist_1, hist_2, cv2.cv.CV_COMP_CHISQR)
elif(method=="bhattacharyya"):
comparison = cv2.compareHist(hist_1, hist_2, cv2.cv.CV_COMP_BHATTACHARYYA)
else:
raise ValueError('[DEEPGAZE] color_classification.py: the method specified ' + str(method) + ' is not supported.')
return comparison
def hist_measure(all_tf, subsampled_tf, plot_fig = False):
sorted_all_tf = sorted(all_tf.items(), key=operator.itemgetter(1)) # becomes tuple
# create subsampled histogram
sub_hist = []
all_hist = []
for item in sorted_all_tf:
all_hist += [item[1]]
if item[0] in subsampled_tf:
sub_hist += [subsampled_tf[item[0]]]
else:
sub_hist += [0]
# correlation
all_array = np.array(all_hist)
sub_array = np.array(sub_hist)
all_array = all_array / (np.sum(all_array) * 1.0)
sub_array = sub_array / (np.sum(sub_array) * 1.0)
all_hist_cv = all_array.ravel().astype('float32')
sub_hist_cv = sub_array.ravel().astype('float32')
correlation_score = cv2.compareHist(all_hist_cv, sub_hist_cv, cv2.cv.CV_COMP_CORREL)
chisq_dist = cv2.compareHist(all_hist_cv, sub_hist_cv, cv2.cv.CV_COMP_CHISQR)
l1_dist = L1_dist(all_array, sub_array)
cos_sim = cos_similarity(all_array, sub_array)
'''
print 'Correlation (higher-> similar) :', cv2.compareHist(all_hist_cv, sub_hist_cv, cv2.cv.CV_COMP_CORREL)
print 'Chi-square (smaller-> similar) :', cv2.compareHist(all_hist_cv, sub_hist_cv, cv2.cv.CV_COMP_CHISQR)
print 'L1 distance (smaller -> similar):', L1_dist(all_array, sub_array)
'''
if plot_fig:
ax = plt.subplot(2,1,1)
plt.plot(range(len(all_hist)), all_array, 'o-r', label = 'Nonsubsampled')
plt.plot(range(len(sub_hist)), sub_array, 'x-b', label = 'Subsampled')
plt.plot(range(len(all_hist)), abs(all_array - sub_array), 'o-k', label= 'L1 dist')
plt.legend()
plt.xlabel('Word Index')
plt.ylabel('Word Count (#)')
ax.set_title('Histogram')
ax = plt.subplot(2,1,2)
plt.plot(sub_hist, all_hist, 'o')
plt.xlabel('Subsampled Word Count (#)')
plt.ylabel('Nonsubsampled Word Count (#)')
ax.set_title('Correlation')
plt.show()
return l1_dist, correlation_score, chisq_dist, cos_sim
def get_shots(video, threshold):
if len(video) == 0:
return [], []
if len(video) == 1:
return [video], []
shots = [Video(frames = [video[0]])]
function = []
for i in xrange(1, len(video)):
previous = video[i - 1].getHistogram()
current = video[i].getHistogram()
t = 0
for ti in range(len(current)):
t += cv2.compareHist(previous[ti], current[ti], cv.CV_COMP_CORREL)
t = t / len(current)
function.append(t)
if t < threshold:
shots.append(Video(frames = [video[i]]))
else:
shots[-1].addFrame(video[i])
return shots, function
def object_comparisson(self, roi):
roi_hist = self.calculate_histogram(roi)
# Tomo el histograma del objeto para comparar
obj_hist = self.saved_object_comparisson()
return cv2.compareHist(roi_hist, obj_hist, cv2.cv.CV_COMP_BHATTACHARYYA)
def __eq__(self,another_histo):
difference=cv2.compareHist(self.hist[0],another_histo.hist[0],cv2.cv.CV_COMP_CORREL)
difference*=100
if difference>=__THRESH_VAL__:
return True
return False
def similarness(image1,image2):
"""
Return the correlation distance be1tween the histograms. This is 'normalized' so that
1 is a perfect match while -1 is a complete mismatch and 0 is no match.
"""
# Open and resize images to 200x200
i1 = Image.open(image1).resize((200,200))
i2 = Image.open(image2).resize((200,200))
# Get histogram and seperate into RGB channels
i1hist = numpy.array(i1.histogram()).astype('float32')
i1r, i1b, i1g = i1hist[0:256], i1hist[256:256*2], i1hist[256*2:]
# Re bin the histogram from 256 bins to 48 for each channel
i1rh = numpy.array([sum(i1r[i*16:16*(i+1)]) for i in range(16)]).astype('float32')
i1bh = numpy.array([sum(i1b[i*16:16*(i+1)]) for i in range(16)]).astype('float32')
i1gh = numpy.array([sum(i1g[i*16:16*(i+1)]) for i in range(16)]).astype('float32')
# Combine all the channels back into one array
i1histbin = numpy.ravel([i1rh, i1bh, i1gh]).astype('float32')
# Same steps for the second image
i2hist = numpy.array(i2.histogram()).astype('float32')
i2r, i2b, i2g = i2hist[0:256], i2hist[256:256*2], i2hist[256*2:]
i2rh = numpy.array([sum(i2r[i*16:16*(i+1)]) for i in range(16)]).astype('float32')
i2bh = numpy.array([sum(i2b[i*16:16*(i+1)]) for i in range(16)]).astype('float32')
i2gh = numpy.array([sum(i2g[i*16:16*(i+1)]) for i in range(16)]).astype('float32')
i2histbin = numpy.ravel([i2rh, i2bh, i2gh]).astype('float32')
return cv2.compareHist(i1histbin, i2histbin, 0)
def get_post(shots, n):
px = 10
py = 10
threshold = 0.4
nb_parts = 30
for i in range(2, len(shots)):
first = shots[i][0].getPartHistograms(px,py)
for m in range(n):
j = i - 2 - n
if j < 0:
break
last = shots[j][-1].getPartHistograms(px, py)
t = 0
for ti in range(len(first)):
v = cv2.compareHist(last[ti], first[ti], cv.CV_COMP_CORREL)
if v < threshold:
t += 1
if t > nb_parts:
if len(shots[i]) > 1:
shots[i] = shots[i][1:]
else:
if len(shots[j]) == 1:
shots[i] = []
break
return shots
def extraerVecinos(imagen1,capitales):
listaImagenes=[]
try:
imagenes = devolverImagenesCapitales(imagen1,capitales)
img1 = cv2.imread( 'Fotos/'+imagen1 );
v = cv2.calcHist([img1], [0, 1, 2], None, [8, 8, 8], [0, 256, 0, 256, 0, 256]) #Calculo de histograma de imagen
v = v.flatten()
hist1 = v / sum(v)
dictSumas ={}
for imagen2 in imagenes:
if not imagen2==imagen1:
try:
img2 = cv2.imread( 'Fotos/'+imagen2);
v = cv2.calcHist([img2], [0, 1, 2], None, [8, 8, 8], [0, 256, 0, 256, 0, 256]) #Calculo de histograma de imagen
v = v.flatten()
hist2 = v / sum(v)
d = cv2.compareHist( hist1, hist2, cv2.cv.CV_COMP_INTERSECT) #Calculo de similitud de imagenes
dictSumas[imagen2] = d
except:
print "Error"
dictSumas = dictSumas.items()
dictSumas.sort(lambda x,y:cmp(y[1], x[1]))
i=0
while i<K: #Devuelve las K imagenes mas similares.
listaImagenes.append(dictSumas[i][0])
i=i+1
except:
print "Error"
return listaImagenes
def calculate_weights(self):
"""Iterates over x and y starting at the frame radius by dx and dy
until the window size. At each 'coordinate', computes an object around
it, computes a histogram, and calculates a weight by comparing that
histogram to the original frame. Subtracts the normalized distance
from the object of the original frame, and creates a list of all of
these new weights, saved as a numpy array.
"""
weights = []
for x in range(frame.rad, frame.window_x, frame.dx):
for y in range(frame.rad, frame.window_y, frame.dy):
obj = new_frame.create_general_object(x,y)
hist = new_frame.create_general_hist(obj)
# compare histograms to find weight
weight = cv2.compareHist(frame.hist, hist, method=cv2.cv.CV_COMP_CORREL)
# find distance away from old point, and normalize by max distance
max_distance = float(self.find_hypotenuse(frame.window_x, frame.window_y))
distance = self.find_hypotenuse(x-frame.x, y-frame.y) / max_distance
# subtract normalized distance from weight
weight = weight - distance
# make sure no weights are negative
if weight < 0:
weight = 0
# append weights to array
weights.append(weight)
self.weights = np.array(weights)
def run(query, k=10, mode='bruteforce'):
if query.startswith('['):
vec = json.loads(query)
vec = np.array(vec, dtype=np.float32).reshape(1, 512)
elif query in image_map:
vec = image_map[query].reshape(1, 512)
else:
return {'error': 'Could not find data for image: ' + query}
images = []
if mode == 'bruteforce':
hist = vec.reshape(512)
dists = []
for (i, file) in enumerate(image_files):
dist = cv2.compareHist(hist, image_map[file], cv2.cv.CV_COMP_INTERSECT)
dists.append((dist, i))
top = sorted(dists, reverse=True)[:int(k)]
for distance, index in top:
images.append({
'id': image_files[index],
'features': image_map[image_files[index]].tolist(),
'distance': distance
})
return images
def __get_uniq_faces_curr_frame(self, frame_id, faces_roi_hists_prev, faces_roi_hists):
faces_prev = len(faces_roi_hists_prev)
faces_curr = len(faces_roi_hists)
logger.info("[{0}] Face Similarity: Prev: {1}, Curr: {2}".format(frame_id, faces_prev, faces_curr))
# First Time
if faces_prev == 0:
return faces_curr
# Current frame has more faces than prev frame
# if faces_curr > faces_prev:
# return faces_curr - faces_prev
uniq_faces_curr_frame = 0
# Perform Image Histogram Similarity
# For each histogram in current frame
for rh1 in faces_roi_hists:
match_found = False
# For each histogram in previous frame
for rh2 in faces_roi_hists_prev:
# print "\nrh1 {0}: {1}".format(type(rh1),np.shape(rh1))
# print "\nrh2 {0}: {1}".format(type(rh2),np.shape(rh2))
corr = cv2.compareHist(rh1, rh2, cv2.HISTCMP_CORREL)
logger.info("[{0}] Similarity Metrics: {1}".format(frame_id, corr))
if corr >= 0.35:
# Match Found, can break the loop for this histogram in current frame
match_found = True
break
# Add to unique face count, if no match found for this histogram in current frame
if not match_found:
uniq_faces_curr_frame += 1
logger.info("[{0}] Total Unique Faces in Current Frame: {1}".format(frame_id, uniq_faces_curr_frame))
return uniq_faces_curr_frame
def similarity(self, f1,f2): totalD=0 for hist1,hist2 in zip(f1,f2): totalD+=cv2.compareHist(hist1,hist2,0 ) return totalD
def sliding_window(image,r1,r2,step,roihist):
test_path = "/home/sarbajit/PyCharm_Scripts/test/green_pad_same_name_new/final_rotated2/"
item = image
if item.endswith(".png") or item.endswith(".PNG"):
x = test_path+item
target2 = cv2.imread(x)
target = target2[90:150, 90:520]
(winW, winH) = (50, 30)
for (x, y, window) in sliding_window_test(target,r1,r2,stepSize=step, windowSize=(winW, winH)):
# if the window does not meet our desired window size, ignore it
if window.shape[0] != winH or window.shape[1] != winW:
continue
#this section does the histogram backprojected matching window by window.
hsvt = cv2.cvtColor(window, cv2.COLOR_BGR2HSV)
inputImage = cv2.calcHist([hsvt], [0, 1], None, [180, 256], [0, 180, 0, 256])
cv2.normalize(roihist, roihist, 0, 255, cv2.NORM_MINMAX)
dst = cv2.calcBackProject([hsvt], [0, 1], roihist, [0, 180, 0, 256], 1)
match = cv2.compareHist(roihist, inputImage, method=0)
print match
#the match is printed to see the difference and jumps when the window moves through the landing pad
# THIS IS WHERE YOU WOULD PROCESS YOUR WINDOW,AND DO THE NECESSARY STEPS
# we'll just draw the window and show the results
clone = target.copy()
cv2.rectangle(clone, (x, y), (x + winW, y + winH), (0, 255, 0), 2)
cv2.imshow("window", clone)
cv2.waitKey(1)
time.sleep(1)
# sliding_window('2015-08-06_06-27-48.png',28,58,30)
def main():
dict_path = TEMP_DIR + sys.argv[1]
dictionary = pickle.load(open(dict_path, 'rb'))
index_path = TEMP_DIR + sys.argv[2]
index = pickle.load(open(index_path, 'rb'))
image_path = IMAGES_DIR + sys.argv[3]
image_des = describe(image_path, dictionary)
result_path = TEMP_DIR + sys.argv[4]
clear_directory(result_path)
result = []
for path, des in index.items():
likelyhood = cv2.compareHist(image_des, des, cv2.cv.CV_COMP_CHISQR)
result.append((likelyhood, path))
best_match = sorted(result, key=lambda x: x[0])[:NUM_MATCHES]
for rank, i in enumerate(best_match):
likelyhood, path = i
os.symlink(path, result_path + '/{}.jpeg'.format(rank))
pprint.pprint(best_match)
def sort_hist_dissim(self):
""" Sort by histigram of face dissimilarity """
input_dir = self.args.input_dir
logger.info("Sorting by histogram dissimilarity...")
img_list = [
[img,
cv2.calcHist([cv2.imread(img)], [0], None, [256], [0, 256]), 0]
for img in
tqdm(self.find_images(input_dir), desc="Loading", file=sys.stdout)
]
img_list_len = len(img_list)
for i in tqdm(range(0, img_list_len), desc="Sorting", file=sys.stdout):
score_total = 0
for j in range(0, img_list_len):
if i == j:
continue
score_total += cv2.compareHist(img_list[i][1],
img_list[j][1],
cv2.HISTCMP_BHATTACHARYYA)
img_list[i][2] = score_total
logger.info("Sorting...")
img_list = sorted(img_list, key=operator.itemgetter(2), reverse=True)
return img_list
def match_histograms(query_histogram, location_histogram, weights=None):
"""
Match query histogram with location histogram and return the
distance. To do this, it needs a distance measurement.
"""
# TODO: I could use the distance function as a parameter
#dist = np.linalg.norm(query_histogram - location_histogram)
#dist = spatial.distance.cosine(query_histogram, location_histogram)
dist = cv2.compareHist(np.float32(query_histogram), np.float32(location_histogram), 4)
#dist = JSD(np.float32(query_histogram), np.float32(location_histogram))
#_, dist = stats.ks_2samp(query_histogram, location_histogram)
#dist = -dist
# dist = cv2.EMD(np.float32(query_histogram), np.float32(location_histogram), 3)
#dist= stats.entropy(np.float32(query_histogram), np.float32(location_histogram))
#f = np.float64([1] * len(query_histogram))
#s = np.float64([1] * len(query_histogram))
#dist = emd.emd(np.float64(query_histogram), np.float64(location_histogram), f, s)
return dist
def hist_comp(imlist, loadfunc=None, method="correlation"):
"""
Histogram comparison
:param imlist: list of path to images or arrays
:return: comparison
"""
# http://www.pyimagesearch.com/2014/07/14/3-ways-compare-histograms-using-opencv-python/
#image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
# assert(len(fns)>=2) # no images to compare. There must be 2 or more
def helper(im, loadfunc=loadfunc):
if isinstance(im, basestring):
if loadfunc is None:
def loadfunc(im):
return cv2.imread(im)
im = loadfunc(im) # read BGR image
return im
method, reverse = hist_map[method]
comp, comparison = None, []
for i, im in enumerate(imlist): # for each image get data
hist = cv2.calcHist([helper(im)], [0, 1, 2], None, [
8, 8, 8], [0, 256, 0, 256, 0, 256])
hist = cv2.normalize(hist).flatten()
if i == 0:
comp = hist
comparison.append((cv2.compareHist(comp, hist, method), im))
comparison.sort(key=lambda x: x[0], reverse=reverse) # sort comparisons
return comparison
def sort_hist(self):
""" Sort by histogram of face similarity """
input_dir = self.args.input_dir
logger.info("Sorting by histogram similarity...")
img_list = [
[img, cv2.calcHist([cv2.imread(img)], [0], None, [256], [0, 256])]
for img in
tqdm(self.find_images(input_dir), desc="Loading", file=sys.stdout)
]
img_list_len = len(img_list)
for i in tqdm(range(0, img_list_len - 1), desc="Sorting",
file=sys.stdout):
min_score = float("inf")
j_min_score = i + 1
for j in range(i + 1, len(img_list)):
score = cv2.compareHist(img_list[i][1],
img_list[j][1],
cv2.HISTCMP_BHATTACHARYYA)
if score < min_score:
min_score = score
j_min_score = j
(img_list[i + 1],
img_list[j_min_score]) = (img_list[j_min_score],
img_list[i + 1])
return img_list
def Harris_Corner(self):
self.threshold = 0.999999999999
temp_i = self.image_i.copy()
temp1_i = self.image_i.copy()
gray_i = cv2.cvtColor(temp_i, cv2.COLOR_BGR2GRAY)
gray_i = numpy.float32(gray_i)
dst_i = cv2.cornerHarris(gray_i, 2, 3, 0.025)
dst_i = cv2.dilate(dst_i, None)
# Threshold for an optimal value, it may vary depending on the image.
temp_i[dst_i < 0.01 * dst_i.max()] = [0, 0, 0]
temp1_i[dst_i > 0.01 * dst_i.max()] = [0, 0, 255]
hist_i = cv2.calcHist([temp_i], [0], None, [256], [0, 256])
temp_j = self.image_j.copy()
temp1_j = self.image_j.copy()
gray_j = cv2.cvtColor(temp_j, cv2.COLOR_BGR2GRAY)
gray_j = numpy.float32(gray_j)
dst_j = cv2.cornerHarris(gray_j, 2, 3, 0.025)
dst_j = cv2.dilate(dst_j, None)
# Threshold for an optimal value, it may vary depending on the image.
temp_j[dst_j < 0.01 * dst_j.max()] = [0, 0, 0]
temp1_j[dst_j > 0.01 * dst_j.max()] = [0, 0, 255]
hist_j = cv2.calcHist([temp_j], [0], None, [256], [0, 256])
self.measure = cv2.compareHist(hist_i, hist_j, cv.CV_COMP_CORREL)
self.assertGreater(self.measure, self.threshold)
print self.measure
def histogram_compare(img, params):
resolution = 100
h1 = cv2.calcHist([img], [0], params.fg_mask, [resolution], [0, 1.0])
h2 = cv2.calcHist([img], [0], params.bg_mask, [resolution], [0, 1.0])
cv2.normalize(h1, h1, alpha=1, norm_type=1)
cv2.normalize(h2, h2, alpha=1, norm_type=1)
return cv2.compareHist(h1, h2, 0)
def show_ten_quantized_closest(frame_data,frame_block_dict,target_frame_number, description ):
target_frame_block_hist_dict = frame_block_dict[target_frame_number]
top_ten_frames = list()
print("Comparing frames...")
for keyA in frame_block_dict:
if keyA == target_frame_number: #dont compare the frame against itself
continue
else:
frame_score = float(0)
for keyB in frame_block_dict[keyA]:
block_hist = frame_block_dict[keyA][keyB[0],keyB[1]]
frame_score += cv2.compareHist(target_frame_block_hist_dict[keyB[0],keyB[1]],block_hist, 2) #Intersection compare
top_ten_frames.append((keyA, frame_score))
top_ten_frames.sort(key=lambda tup: tup[1]) # sorts in place
top_ten_frames.reverse()
top_ten_frame_indexes = list((x[0] for x in top_ten_frames)) #just need to the frame number, not the diff so we strip that out
top_ten_frame_values = list((x[1] for x in top_ten_frames))
for i in range(0,10):
# For each of those frame numbers, display the image in a window with the number
index = top_ten_frame_indexes[i]
score = top_ten_frame_values[i]
rgb_target = cv2.cvtColor(frame_data[index-1].astype(np.uint8), cv2.COLOR_GRAY2BGR)
frame_description = description + ' #' + str(i + 1) + ': frame ' + str(index) + '. Score: ' + str(score)
print frame_description
cv2.imshow(frame_description, rgb_target)
cv2.waitKey(0)
cv2.destroyAllWindows()
return
def do_comparison(self, out):
alsum = sorted(self.method_composites['alSum'].items(), key=lambda x: x[0])
random = sorted(self.method_composites['random'].items(), key=lambda x: x[0])
temporal = sorted(self.method_composites['temporalInterval'].items(), key=lambda x: x[0])
for ((asite, acomp), (rsite, rcomp)), (tsite, tcomp) in zip(zip(alsum, random), temporal):
print(acomp, rcomp, tcomp)
aHist = gethistogram(cv2.imread(self.path + acomp, cv2.IMREAD_COLOR))
rHist = gethistogram(cv2.imread(self.path + rcomp, cv2.IMREAD_COLOR))
tHist = gethistogram(cv2.imread(self.path + tcomp, cv2.IMREAD_COLOR))
arSim = math.fabs(cv2.compareHist(aHist, rHist, cv2.HISTCMP_CORREL))
atSim = math.fabs(cv2.compareHist(aHist, tHist, cv2.HISTCMP_CORREL))
out.write("%s,%f,%f\n" % (asite, arSim, atSim))
del aHist
del rHist
del tHist
def Compare_similarity(self, refBox, cropBox): # Compute the similarity between two ROI # Param refBox : ROI for reference # Param cropBox : ROI for comparing H1 = cv2.normalize(cv2.calcHist(refBox, [0, 1, 2], None, [8, 8, 8], [0, 256, 0, 256, 0, 256])).flatten() H2 = cv2.normalize(cv2.calcHist(cropBox, [0, 1, 2], None, [8, 8, 8], [0, 256, 0, 256, 0, 256])).flatten() return cv2.compareHist(H1, H2, cv2.cv.CV_COMP_CORREL)
def getMatches():
init()
results = {}
reverse = False
# # if we are using the correlation or intersection
# # method, then sort the results in reverse order
# if methodName in ("Correlation", "Intersection"):
# reverse = True
# loop over the index
for (k, hist) in index.items():
# compute the distance between the two histograms
# using the method and update the results dictionary
d = cv2.compareHist(index["doge.png"], hist, method)
results[k] = d
# sort the results
results = sorted([(v, k) for (k, v) in results.items()], reverse=reverse)
print("the method is :----" + methodName)
sorted_list = []
# loop over the results
for (i, (v, k)) in enumerate(results):
# show the result
sorted_list.append(str(k))
# print sorted_list(i)
print("name : " + str(k) + " value :" + str(v))
return sorted_list[2:]
def matches(self,frame,rects):
cal=lambda x:Match.calHist(frame,x)
hists=list(map(cal,rects))
print(list(map(lambda x:cv2.compareHist(x,self.hist,cv2.HISTCMP_BHATTACHARYYA),hists)))
def main():
"""
Description:
----------
Parameter:
----------
* None
Return:
----------
* None
"""
# ===== Standart Histgram Data 読み込み===== #
_img_hist_stand_red = cv2.imread("./hist_standard/R-1.jpg", 1)
_img_hist_stand_red_hsv = cv2.cvtColor(_img_hist_stand_red,
cv2.COLOR_BGR2HSV)
_img_hist_stand_red_h = cv2.calcHist(_img_hist_stand_red_hsv, [0], None,
[180], [0, 180])
_img_hist_stand_red_h_w, _img_hist_stand_red_h_h, _img_hist_stand_red_h_ch = _img_hist_stand_red.shape[:
3]
# plt.hist(_img_hist_stand_red_h.ravel(), 256, [0, 256])
# plt.hist(_img_hist_stand_red_h, 256, [0, 256])
# plt.show()
# cv2.namedWindow("hist", cv2.WINDOW_NORMAL)
# cv2.imshow("hist", _img_hist_stand_red)
# ===== 画像読み込み ===== #
_img_src = cv2.imread("./img_data/temp1.jpg", 1)
print("size={}".format(_img_src.shape))
# 画像リサイズ
_img_src = cv2.resize(_img_src, (1280, 720))
while True:
time_start = time.time()
# cv2.namedWindow("_img_src", cv2.WINDOW_NORMAL)
# cv2.imshow("_img_src", _img_src)
# ===== グレイスケールへ変換 ===== #
"""
一定しきい値を超える明るさの領域を探す
"""
# BGR色空間からGrayへ変換
_img_gray = cv2.cvtColor(_img_src, cv2.COLOR_BGR2GRAY)
# cv2.namedWindow("_img_gray", cv2.WINDOW_NORMAL)
# cv2.imshow("_img_gray", _img_gray)
ret, _img_thresh = cv2.threshold(_img_gray, 220, 255,
cv2.THRESH_TOZERO)
# cv2.namedWindow("_img_thresh", cv2.WINDOW_NORMAL)
# cv2.imshow("_img_thresh", _img_thresh)
# ===== Morphology ===== #
_ERODE_KERNEL = np.ones((3, 3), np.uint8)
"""
_img_erode_dst = cv2.erode(img_thresh, _ERODE_KERNEL, iterations = 1)
cv2.namedWindow("_img_erode_dst", cv2.WINDOW_NORMAL)
cv2.imshow("_img_erode_dst", _img_erode_dst)
"""
_img_opening_dst = cv2.morphologyEx(_img_thresh, cv2.MORPH_OPEN,
_ERODE_KERNEL)
cv2.namedWindow("_img_opening", cv2.WINDOW_NORMAL)
cv2.imshow("_img_opening", _img_opening_dst)
# ===== Labering処理 ===== #
"""
- 重心座標
- サイズ
"""
_debug_src = _img_src.copy()
num, _data = ELP.expansion_labeling_prcessing(_img_opening_dst,
[50, 2000], [0.0, 1.0],
[False, False],
_debug_src)
print(num)
# ===== HistgramInterseption ===== #
"""
- Labering処理で得た明るい領域の位置をトリミングする.
- 各領域において, 敵チームの色のヒストグラムとベースヒスグラムデータを照らしあわせて比較.
"""
img_trim = _img_src.copy()
img_hist_dst = _img_src.copy()
img_armer_trim = [0 for i in range(0, num)]
# print(img_armer_trim)
_hist_object = np.zeros((num, 1), dtype=np.float64)
hist_armer_trim = [0 for i in range(0, num)]
for i in range(0, num):
# triming
img_armer_trim[i] = img_trim[int(_data[i][5]):int(_data[i][5] +
_data[i][7]),
int(_data[i][4]):int(_data[i][4] +
_data[i][6])]
# resize
img_armer_trim[i] = cv2.resize(
img_armer_trim[i],
(_img_hist_stand_red_h_w, _img_hist_stand_red_h_h))
# print("hist w={}, h={}".format(_img_hist_stand_red_h_w, _img_hist_stand_red_h_h))
# print("hist w={}, h={}".format(_img_hist_stand_red_h_w, _img_hist_stand_red_h_h))
# convert to hsv
img_armer_trim[i] = cv2.cvtColor(img_armer_trim[i],
cv2.COLOR_BGR2HSV)
# get hist
hist_armer_trim[i] = cv2.calcHist(img_armer_trim[i], [0], None,
[180], [0, 180])
# comper hist
_hist_object[i] = cv2.compareHist(hist_armer_trim[i],
_img_hist_stand_red_h,
cv2.HISTCMP_CORREL)
print("detected object{}={}".format(i, _hist_object[i]))
if (0.05 <= _hist_object[i]):
img_hist_dst = cv2.rectangle(
img_hist_dst, (int(_data[i][4]), int(_data[i][5])),
(int(_data[i][4] + _data[i][6]),
int(_data[i][5] + _data[i][7])), (0, 255, 0), 3)
cv2.namedWindow("img_hist_dst", cv2.WINDOW_NORMAL)
cv2.imshow("img_hist_dst", img_hist_dst)
# ===== 装甲板を狙う処理 =====#
"""
- 装甲板についてる両側のLEDを見つけるため, 同じ形(アスペクト比とか)かつ, 一定ピクセル距離のものを見つける.
- 条件を満たすものであれば, 真ん中の重心座標を狙う.
"""
# ===== 相手の機体番号の取得 ===== #
"""
- 装甲板の場所がわかれば, 装甲板のみをトリミングして, 番号を取得
取得方法は, OCRか, テンプレートマッチングか(多分後者のほうが早そう)
"""
# time.sleep(1.0)
time_end = time.time()
print("[ProcessTime] {}[s], {}[FPS]".format(
time_end - time_start, 1 / (time_end - time_start)))
# ===== key Event ===== #
_key = cv2.waitKey(1)
if (_key):
if (_key == ord("q")):
print("finish")
break
cv2.destroyAllWindows()
def upload():
shutil.rmtree(SAVE_DIR)
os.mkdir(SAVE_DIR)
# if request.method == 'POST':
# name = request.form.get('names')
name = request.form['names']
name = str(name)
if name == "beige":
SUB_DIR = 'beige/'
elif name == "black":
SUB_DIR = 'black/'
elif name == 'blue':
SUB_DIR = 'blue/'
elif name == 'brown':
SUB_DIR = 'brown/'
elif name == "check":
SUB_DIR = 'check/'
elif name == "gray":
SUB_DIR = 'gray/'
elif name == "green":
SUB_DIR = 'green/'
elif name == "multi_tone":
SUB_DIR = 'multi_tone/'
elif name == "pink":
SUB_DIR = 'pink/'
elif name == "red":
SUB_DIR = 'red/'
elif name == "Polka_dot":
SUB_DIR = 'Polka_dot/'
elif name == "yellow":
SUB_DIR = 'yellow/'
elif name == "actress":
SUB_DIR = 'actress/'
else:
SUB_DIR = 'unicro/'
# # ファイルがなかった場合の処理
# if 'file' not in request.files:
# flash('ファイルがありません','failed')
# return redirect(request.url)
# file = request.files['image']
# # ファイルのチェック
# if file and allowed_file(file.filename):
# 危険な文字を削除（サニタイズ処理）
# filename = secure_filename(file.filename)
# 画像として読み込み
stream = request.files['image'].stream
img_array = np.asarray(bytearray(stream.read()), dtype=np.uint8)
img = cv2.imdecode(img_array, 1)
img_size = (200, 200)
channels = (0, 1, 2)
mask = None
hist_size = 256
ranges = (0, 256)
ret = {}
if SUB_DIR != 'actress/':
target_img = img
target_img = cv2.resize(target_img, img_size)
# if target_img and allowed_file(target_img.filename):
# filename = secure_filename(target_img.filename)
# target_img.save(os.path.join('/uploads', filename))
# img_url = '/uploads/' + filename
comparing_files = os.listdir(IMG_DIR + SUB_DIR)
if len(comparing_files) == 0:
sys.exit(1)
for comparing_file in comparing_files:
if comparing_file == '.DS_Store':
continue
tmp = []
if not comparing_file.endswith(('.png', '.jpg', '.jpeg')):
continue
for channel in channels:
target_hist = cv2.calcHist([target_img], [channel], mask,
[hist_size], ranges)
comparing_img = cv2.imread(IMG_DIR + SUB_DIR + comparing_file)
comparing_img = cv2.resize(comparing_img, img_size)
# calc hist of comparing image
comparing_hist = cv2.calcHist([comparing_img], [channel], mask,
[hist_size], ranges)
# compare hist
tmp.append(cv2.compareHist(target_hist, comparing_hist, 0))
# mean hist
ret[comparing_file] = mean(tmp)
# sort
#####################################3
if SUB_DIR == 'actress/':
# if img and allowed_file(img.filename):
# filename = secure_filename(img.filename)
# img.save(os.path.join('/uploads', filename))
# img_url = '/uploads/' + filename
target_img = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
target_img = cv2.resize(target_img, img_size)
bf = cv2.BFMatcher(cv2.NORM_HAMMING)
# detector = cv2.ORB_create()
detector = cv2.AKAZE_create()
(_, target_des) = detector.detectAndCompute(target_img, None)
comparing_files = os.listdir(IMG_DIR + SUB_DIR)
for comparing_file in comparing_files:
if comparing_file == '.DS_Store':
continue
comparing_img_path = IMG_DIR + SUB_DIR + comparing_file
try:
comparing_img = cv2.imread(comparing_img_path,
cv2.IMREAD_GRAYSCALE)
comparing_img = cv2.resize(comparing_img, img_size)
(_, comparing_des) = detector.detectAndCompute(
comparing_img, None)
matches = bf.match(target_des, comparing_des)
dist = [m.distance for m in matches]
score = sum(dist) / len(dist)
if score <= 1:
score = 1
score = 100.0 / score
except cv2.error:
score = 100000
ret[comparing_file] = score
############################################################
dic_sorted = sorted(ret.items(), reverse=True, key=lambda x: x[1])[:3]
estimated_d = []
for file in dic_sorted:
img_path = IMG_DIR + SUB_DIR + file[0]
img = cv2.imread(img_path)
# cv2.imshow('image',img)
# 保存
dt_now = datetime.now().strftime("%Y_%m_%d%_H_%M_%S_%f")
save_path = os.path.join(SAVE_DIR, dt_now + ".jpeg")
cv2.imwrite(save_path, img)
estimated_d.append(file[1])
f_imgs = os.listdir(SAVE_DIR)
if '.DS_Store' in f_imgs:
f_imgs.remove('.DS_Store')
exists_img = sorted(f_imgs)[-3:]
# ファイルの保存
# file.save(os.path.join(app.config['UPLOAD_FOLDER'], filename))
# アップロード後のページに転送
return render_template('index.html',
names=names,
data=zip(exists_img, estimated_d))
channels,
None,
histSize,
ranges,
accumulate=False)
cv.normalize(hist_test, hist_test, alpha=0, beta=1, norm_type=cv.NORM_MINMAX)
# 1 Correlation
# 2 Chi-Square
# 3 Intersection
# 4 Bhattacharyya distance
for compare_method in range(4):
comparisons = []
for hist in images_hist:
comp = cv.compareHist(hist_test, hist, compare_method)
comparisons.append(comp)
if compare_method == 3:
should_rev = False
else:
should_rev = True
matches = [
im for (c, im) in sorted(zip(comparisons, images),
key=lambda pair: pair[0],
reverse=should_rev)
]
for i in range(2):
im = matches[i]
def compare_two_hist(hist1, hist2, md="Correlation"):
if md not in OPENCV_METHODS:
md = "Correlation"
d = cv2.compareHist(hist1, hist2, OPENCV_METHODS[md])
return d
def to_dict_w_hists(data_dict, keys, data_zip):
i = 0
while i < len(keys):
data_dict[keys[i]]['name'] = data_zip[i][0]
data_dict[keys[i]]['arrays'] = {}
data_dict[keys[i]]['arrays']['full'] = {}
data_dict[keys[i]]['arrays']['full']['array'] = data_zip[i][1]
data_dict[keys[i]]['arrays']['full']['numpy hist'] = np.histogram(
data_zip[i][1])
data_dict[keys[i]]['arrays']['full']['cv2 hist'] = np_hist_to_cv(
np.histogram(data_zip[i][1]))
data_dict[keys[i]]['MSE'] = round(data_zip[i][2], 2)
data_dict[keys[i]]['SSIM'] = round(data_zip[i][3], 2)
data_dict[keys[i]]['arrays']['top left'] = {}
data_dict[keys[i]]['arrays']['top left']['array'] = data_zip[i][4]
data_dict[keys[i]]['arrays']['top left']['numpy hist'] = np.histogram(
data_zip[i][4])
data_dict[keys[i]]['arrays']['top left']['cv2 hist'] = np_hist_to_cv(
np.histogram(data_zip[i][4]))
data_dict[keys[i]]['arrays']['top right'] = {}
data_dict[keys[i]]['arrays']['top right']['array'] = data_zip[i][5]
data_dict[keys[i]]['arrays']['top right']['numpy hist'] = np.histogram(
data_zip[i][5])
data_dict[keys[i]]['arrays']['top right']['cv2 hist'] = np_hist_to_cv(
np.histogram(data_zip[i][5]))
data_dict[keys[i]]['arrays']['low left'] = {}
data_dict[keys[i]]['arrays']['low left']['array'] = data_zip[i][6]
data_dict[keys[i]]['arrays']['low left']['numpy hist'] = np.histogram(
data_zip[i][6])
data_dict[keys[i]]['arrays']['low left']['cv2 hist'] = np_hist_to_cv(
np.histogram(data_zip[i][6]))
data_dict[keys[i]]['arrays']['low right'] = {}
data_dict[keys[i]]['arrays']['low right']['array'] = data_zip[i][7]
data_dict[keys[i]]['arrays']['low right']['numpy hist'] = np.histogram(
data_zip[i][7])
data_dict[keys[i]]['arrays']['low right']['cv2 hist'] = np_hist_to_cv(
np.histogram(data_zip[i][7]))
data_dict[keys[i]]['IMSE'] = round(data_zip[i][8], 2)
data_dict[keys[i]]['IMSE Map'] = data_zip[i][9]
data_dict[keys[i]]['MSE Map'] = data_zip[i][10]
data_dict[keys[i]]['SSIM Map'] = data_zip[i][11]
data_dict[keys[i]]['CW SSIM'] = data_zip[i][12]
data_dict[keys[i]]['CW SSIM Map'] = data_zip[i][13]
data_dict[keys[i]]['Mag. Map'] = data_zip[i][14]
data_dict[keys[i]]['Frequency Transects'] = data_zip[i][15]
data_dict[keys[i]]['DCT Map'] = data_zip[i][16]
data_dict[keys[i]]['DCT Curve'] = data_zip[i][17]
# Histogram Comparisons
# Bhattacharyya
data_dict[keys[i]]['Bhattacharyya Full'] = round(
cv2.compareHist(data_dict[keys[i]]['arrays']['full']['cv2 hist'],
data_dict[keys[0]]['arrays']['full']['cv2 hist'],
cv2.cv.CV_COMP_BHATTACHARYYA), 2)
data_dict[keys[i]]['Bhattacharyya UL'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['top left']['cv2 hist'],
data_dict[keys[0]]['arrays']['top left']['cv2 hist'],
cv2.cv.CV_COMP_BHATTACHARYYA), 2)
data_dict[keys[i]]['Bhattacharyya UR'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['top right']['cv2 hist'],
data_dict[keys[0]]['arrays']['top right']['cv2 hist'],
cv2.cv.CV_COMP_BHATTACHARYYA), 2)
data_dict[keys[i]]['Bhattacharyya LL'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['low left']['cv2 hist'],
data_dict[keys[0]]['arrays']['low left']['cv2 hist'],
cv2.cv.CV_COMP_BHATTACHARYYA), 2)
data_dict[keys[i]]['Bhattacharyya LR'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['low right']['cv2 hist'],
data_dict[keys[0]]['arrays']['low right']['cv2 hist'],
cv2.cv.CV_COMP_BHATTACHARYYA), 2)
# Chi Square
data_dict[keys[i]]['Chi Square Full'] = round(
cv2.compareHist(data_dict[keys[i]]['arrays']['full']['cv2 hist'],
data_dict[keys[0]]['arrays']['full']['cv2 hist'],
cv2.cv.CV_COMP_CHISQR), 2)
data_dict[keys[i]]['Chi Square UL'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['top left']['cv2 hist'],
data_dict[keys[0]]['arrays']['top left']['cv2 hist'],
cv2.cv.CV_COMP_CHISQR), 2)
data_dict[keys[i]]['Chi Square UR'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['top right']['cv2 hist'],
data_dict[keys[0]]['arrays']['top right']['cv2 hist'],
cv2.cv.CV_COMP_CHISQR), 2)
data_dict[keys[i]]['Chi Square LL'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['low left']['cv2 hist'],
data_dict[keys[0]]['arrays']['low left']['cv2 hist'],
cv2.cv.CV_COMP_CHISQR), 2)
data_dict[keys[i]]['Chi Square LR'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['low right']['cv2 hist'],
data_dict[keys[0]]['arrays']['low right']['cv2 hist'],
cv2.cv.CV_COMP_CHISQR), 2)
# Correlation
data_dict[keys[i]]['Correlation Full'] = round(
cv2.compareHist(data_dict[keys[i]]['arrays']['full']['cv2 hist'],
data_dict[keys[0]]['arrays']['full']['cv2 hist'],
cv2.cv.CV_COMP_CORREL), 2)
data_dict[keys[i]]['Correlation UL'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['top left']['cv2 hist'],
data_dict[keys[0]]['arrays']['top left']['cv2 hist'],
cv2.cv.CV_COMP_CORREL), 2)
data_dict[keys[i]]['Correlation UR'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['top right']['cv2 hist'],
data_dict[keys[0]]['arrays']['top right']['cv2 hist'],
cv2.cv.CV_COMP_CORREL), 2)
data_dict[keys[i]]['Correlation LL'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['low left']['cv2 hist'],
data_dict[keys[0]]['arrays']['low left']['cv2 hist'],
cv2.cv.CV_COMP_CORREL), 2)
data_dict[keys[i]]['Correlation LR'] = round(
cv2.compareHist(
data_dict[keys[i]]['arrays']['low right']['cv2 hist'],
data_dict[keys[0]]['arrays']['low right']['cv2 hist'],
cv2.cv.CV_COMP_CORREL), 2)
i = i + 1
def updateHistfaces(self,current_hist):
color = ('b','g','r')
curs=0
if self.choose_face1:
if len(self.histmoyface1)>0:
for rgbHist in self.histmoyface1:
for i,col in enumerate(color):
curs+= cv2.compareHist(current_hist[i], rgbHist[i], cv2.HISTCMP_CORREL)
curs=curs/len(self.histmoyface1)
print("Distance nouvel histo par rapport à histo face1 : ",curs)
if curs>2.8 :
if len(self.histmoyface1)<3:
self.histmoyface1.append(current_hist)
self.index_moyface1+=1
else :
self.histmoyface1[(self.index_moyface1)%3]=current_hist
self.index_moyface1+=1
print("Update face1")
else :
if len(self.histmoyface2)<3:
self.histmoyface2.append(current_hist)
self.index_moyface2+=1
else :
self.histmoyface2[(self.index_moyface2)%3]=current_hist
self.index_moyface2+=1
self.choose_face1=False
print("Update face2")
else :
if len(self.histmoyface1)<3:
self.histmoyface1.append(current_hist)
self.index_moyface1+=1
else :
self.histmoyface2[(self.index_moyface1)%3]=current_hist
self.index_moyface1+=1
print("Update face1")
else :
if len(self.histmoyface2)>0:
for rgbHist in self.histmoyface2:
for i,col in enumerate(color):
curs+= cv2.compareHist(current_hist[i], rgbHist[i], cv2.HISTCMP_CORREL)
curs=curs/len(self.histmoyface2)
print("Distance nouvel histo par rapport à histo face2: ",curs)
if curs>2.8 :
if len(self.histmoyface2)<3:
self.histmoyface2.append(current_hist)
self.index_moyface2+=1
else :
self.histmoyface2[(self.index_moyface2)%3]=current_hist
self.index_moyface2+=1
print("Update face2")
else :
if len(self.histmoyface1)<3:
self.histmoyface1.append(current_hist)
self.index_moyface1+=1
else :
self.histmoyface1[(self.index_moyface1)%3]=current_hist
self.index_moyface1+=1
self.choose_face1=True
print("Update face1")
else :
print("Bug")
Hist = Hist_p
flow = cv2.calcOpticalFlowFarneback(
prvs,
next,
None,
pyr_scale=0.5, # Taux de réduction pyramidal
levels=3, # Nombre de niveaux de la pyramide
winsize=
15, # Taille de fenêtre de lissage (moyenne) des coefficients polynomiaux
iterations=3, # Nb d'itérations par niveau
poly_n=7, # Taille voisinage pour approximation polynomiale
poly_sigma=1.5, # E-T Gaussienne pour calcul dérivées
flags=0)
Hist_p = ComputeHistFlow(flow)
corr.append(cv2.compareHist(Hist, Hist_p,
cv2.HISTCMP_CORREL)) #metrique = correlation
plot_correlation(corr, seuil)
cv2.namedWindow('Video frames')
cv2.moveWindow('Video frames', 500, 0)
cv2.imshow('Video frames', frame2)
k = cv2.waitKey(15) & 0xff
if k == 27:
break
elif k == ord('s'):
cv2.imwrite('Frame_%04d.png' % index, frame2)
cv2.imwrite('Hist%04d.png' % index, Hist_p)
prvs = next
ret, frame2 = cap.read()
def mature_check(image,human_string,draw_mask):
red_flag = 0
if human_string == "broccoli":
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (20, 20))
threshold_low = np.array([20, 30, 20], dtype = "uint8")
threshold_high = np.array([80, 255, 255], dtype = "uint8")
reference_img = cv2.imread("reference/reference1.jpg")
elif human_string == "cabbage":
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (20, 20))
threshold_low = np.array([20, 40, 30], dtype = "uint8")
threshold_high = np.array([70, 255, 255], dtype = "uint8")
reference_img = cv2.imread("reference/reference2.jpg")
elif human_string == "cauliflower":
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (20, 20))
threshold_low = np.array([0, 0, 30], dtype = "uint8")
threshold_high = np.array([180, 50, 255], dtype = "uint8")
reference_img = cv2.imread("reference/reference3.jpg")
elif human_string == "red capsicum":
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (13, 13))
threshold_low = np.array([0, 60, 75], dtype = "uint8")
threshold_high = np.array([10, 255, 255], dtype = "uint8")
reference_img = cv2.imread("reference/reference4.jpg")
red_flag = 1
#process the reference image
reference_blur = cv2.GaussianBlur(reference_img, (15, 15), 2)
reference_hsv = cv2.cvtColor(reference_blur, cv2.COLOR_BGR2HSV)
if red_flag == 1:
mask1 = cv2.inRange(reference_hsv, threshold_low, threshold_high)
#set second set of threshold values
threshold_low = np.array([170, 60, 75], dtype = "uint8")
threshold_high = np.array([180, 255, 255], dtype = "uint8")
#get final mask
mask2 = cv2.inRange(reference_hsv, threshold_low, threshold_high)
reference_mask = cv2.bitwise_or(mask1, mask2, mask = None)
else:
reference_mask = cv2.inRange(reference_hsv, threshold_low, threshold_high)
reference_mask = cv2.morphologyEx(reference_mask, cv2.MORPH_OPEN, kernel)
reference_masked = cv2.bitwise_and(reference_img,reference_img, mask = reference_mask)
#set the mask for input image
input_masked = cv2.bitwise_and(image,image, mask = draw_mask)
hist1 = cv2.calcHist([image], [0,1], draw_mask, [180,256], [0,180, 0,256])
hist2 = cv2.calcHist([reference_img], [0,1], reference_mask, [180,256], [0,180, 0,256])
val = cv2.compareHist(hist1,hist2,cv2.HISTCMP_CORREL)
plot_histogram(reference_img, "Histogram for Reference Image", mask = reference_mask)
plot_histogram(image, "Histogram for Input Image", mask = draw_mask)
if val < 0:
val = 0
return val
cv2.waitKey(0)
index[filename] = hist
OPENCV_METHODS = (("Correlation", cv2.HISTCMP_CORREL),
("Chi-Squared", cv2.HISTCMP_CHISQR), ("Intersection",
cv2.HISTCMP_INTERSECT),
("Hellinger", cv2.HISTCMP_BHATTACHARYYA))
for (methodName, method) in OPENCV_METHODS:
results = {}
reverse = False
if methodName in ("Correlation", "Intersection"):
reverse = True
for (k, hist) in index.items():
d = cv2.compareHist(index["doge.png"], hist, method)
results[k] = d
results = sorted([(v, k) for (k, v) in results.items()], reverse=reverse)
fig = plt.figure("Query")
ax = fig.add_subplot(1, 1, 1)
ax.imshow(images["doge.png"])
plt.axis("off")
fig = plt.figure("Results: %s" % (methodName))
fig.suptitle(methodName, fontsize=20)
for (i, (v, k)) in enumerate(results):
ax = fig.add_subplot(1, len(images), i + 1)
ax.set_title("%s: %.2f" % (k, v))
def main():
# --- ARGUMENT PARSER AND FILEPATHS ---
# Initialise arguent parser
ap = argparse.ArgumentParser()
# Input options for path to images
ap.add_argument("-d", "--directory", help = "Path to directory of images",
required = False, default = "../data/flowers")
# Input option for target image name
ap.add_argument("-t", "--target_img", help = "Filename of the target image",
required = False, default = "image_0001.jpg")
# Extract inputs
args = vars(ap.parse_args())
img_dir = args["directory"]
target_img = args["target_img"]
# --- IMAGE SEARCH ---
# Print message
print(f"\n[INFO] Initialising image search for {target_img} using color histograms.")
# Get filepath to target image
target_path = os.path.join(img_dir, target_img)
# Get file paths to all images in directory
img_paths = get_paths(img_dir)
# Remove target path from all image paths
img_paths.remove(target_path)
# Create empty target data frame for distances
distances_df = pd.DataFrame(columns=["filename", "chisquare_distance"])
# Get histogram of target image
target_hist = get_histogram(target_path)
# Get histogram for all other images and compare to target histogram
for img_path in tqdm(img_paths):
# Get the name of the image
img_name = os.path.split(img_path)[1]
# Get the histogram of the image
img_hist = get_histogram(img_path)
# Calculate the distance of the image by comparing the target and image histogram
distance = round(cv2.compareHist(target_hist, img_hist, cv2.HISTCMP_CHISQR), 2)
# Append filename and distance to dataframe
distances_df = distances_df.append({"filename": img_name,
"chisquare_distance": distance}, ignore_index = True)
# Sort data frame by distance and reset index
distances_df = distances_df.sort_values("chisquare_distance").reset_index(drop=True)
# --- OUTPUT ---
# Prepare output directory
out_dir = os.path.join("..", "out")
if not os.path.exists(out_dir):
os.mkdir(out_dir)
# Save data frame in output directory, using target image for filename
out_df = os.path.join(out_dir, f"{os.path.splitext(target_img)[0]}_hist.csv")
distances_df.to_csv(out_df)
# Save plot with similar images in output directory, using target image for filename
out_plot = os.path.join(out_dir, f"{os.path.splitext(target_img)[0]}_hist_top3.png")
plot_similar(img_dir, target_img, distances_df, out_plot)
# Print message, and print closest image to target image
print(f"\n[INFO] Output is saved in {out_dir}, the closest image to {os.path.splitext(target_img)[0]} is:")
print(distances_df.iloc[0])
x,y,z=frame1.shape
nextimYuv = cv2.cvtColor(frame2,cv2.COLOR_BGR2YUV)
index=1;
imYuv = cv2.cvtColor(frame1,cv2.COLOR_BGR2YUV) # Passage en niveaux de yuv
y=[]
while(ret):
#Calculs histogrammes noir blanc
hist= cv2.calcHist([imYuv], [0], None, [32], [0,255])
hist=cv2.normalize(hist, hist)
hist1= cv2.calcHist([nextimYuv], [0], None, [32], [0,255])
hist1=cv2.normalize(hist1, hist1)
#Correlation entre histogrammes
y.append(cv2.compareHist(hist, hist1,cv2.HISTCMP_CORREL))
#Plot Correlation
fig2, ax2 =plt.subplots()
ax2 = plt.gca()
ax2.plot(y)
plt.xlabel("Index")
plt.ylabel("Corrélation entre frames")
plt.draw()
fig2.canvas.draw()
imgt = np.fromstring(fig2.canvas.tostring_rgb(), dtype=np.uint8,
sep='')
imgt = imgt.reshape(fig2.canvas.get_width_height()[::-1] + (3,))
imgt = cv2.cvtColor(imgt,cv2.COLOR_RGB2BGR)
cv2.imshow("Correlation entre frames",imgt)
def callback_humanlist(self, human_list_msg, cv_image, openpose_depth_img,
camera_transform, time_):
if len(human_list_msg.human_list) == 0:
return
depth_img = openpose_depth_img
frame = cv_image
found_target = False
pos_list = []
for human in human_list_msg.human_list:
# clothes area
main_points = [1, 2, 5, 8]
x = []
y = []
for idx in main_points:
if human.body_key_points_with_prob[idx] != 0:
if human.body_key_points_with_prob[
idx].x != 0 and human.body_key_points_with_prob[
idx].y != 0:
x.append(human.body_key_points_with_prob[idx].x)
y.append(human.body_key_points_with_prob[idx].y)
if len(x) != 4:
continue
#(top_, right_, bottom_, left_) = (int(min(y)), int(max(x)), int(max(y)), int(min(x)))
(top, right, bottom, left) = (int(
(3 * min(y) + max(y)) / 4), int(
(min(x) + 3 * max(x)) / 4), int((min(y) + 3 * max(y)) / 4),
int((3 * min(x) + max(x)) / 4))
#print((top, right, bottom, left))
diagonal_size = sqrt(
pow((right - left), 2) + pow((top - bottom), 2))
print(diagonal_size)
# neglect unvalid small wrong person object from openpose
if diagonal_size < self.valid_size_of_person:
continue
col = int(round((left + right) / 2))
row = int(round((bottom + top) / 2))
clothes_block = frame[top:bottom, left:right]
depth_arr = depth_img[top:bottom, left:right]
cv2.rectangle(frame, (left, bottom), (right, top), (0, 0, 255), 2)
hist = cv2.calcHist([clothes_block], [0, 1, 2], None, [8, 8, 8],
[0, 256, 0, 256, 0, 256])
hist = cv2.normalize(hist, hist).flatten()
# face area
use_face_area = False
face_points = [1, 17, 18]
f_x = []
f_y = []
for idx in face_points:
if human.body_key_points_with_prob[idx] != 0:
if human.body_key_points_with_prob[
idx].x != 0 and human.body_key_points_with_prob[
idx].y != 0:
f_x.append(human.body_key_points_with_prob[idx].x)
f_y.append(human.body_key_points_with_prob[idx].y)
if len(f_x) == 3:
use_face_area = True
pose_transformed = None
f_pose_transformed = None
if use_face_area:
(top, right, bottom, left) = (int(min(f_y)), int(max(f_x)),
int(max(f_y)), int(min(f_x)))
f_col = int(round((left + right) / 2))
f_row = int(round((bottom + top) / 2))
face_block = frame[top:bottom, left:right]
f_depth_arr = depth_img[top:bottom, left:right]
# select the depth located in (self.clothes_depth_arr_position)(ex 2/3), because some obstacle can cover person's clothes in front of person.
depth_list = []
for i in range(0, len(f_depth_arr)):
for j in range(0, len(f_depth_arr[0])):
if f_depth_arr[i][j] != 0:
depth_list.append(f_depth_arr[i][j])
if len(depth_list) == 0:
continue
else:
sorted_depth_list = sorted(depth_list, reverse=False)
selected_idx = int(
len(sorted_depth_list) * self.face_depth_arr_position)
selected_depth = sorted_depth_list[selected_idx]
cv2.rectangle(frame, (left, bottom), (right, top),
(0, 0, 255), 2)
# personPose : [-y, x, z] by camera_link frame unit : mm
# data_to_send.data = rs2.rs2_deproject_pixel_to_point(self.intrinsics, [col, row], selected_depth) # [col, row]
rs2_pose = rs2.rs2_deproject_pixel_to_point(
self.intrinsics, [f_col, f_row],
selected_depth) # [col, row]
pose_stamped = PoseStamped()
pose_stamped.header.frame_id = 'camera_link'
# personPose : [-y, x, z] unit : mm
pose_stamped.pose.position.x = rs2_pose[1] / 1000
pose_stamped.pose.position.y = -rs2_pose[0] / 1000
pose_stamped.pose.position.z = rs2_pose[2] / 1000
pose_stamped.pose.orientation.z = 0.0
pose_stamped.pose.orientation.w = 1.0
pose_stamped.header.stamp = rospy.Time.now()
f_pose_transformed = tf2_geometry_msgs.do_transform_pose(
pose_stamped, camera_transform)
pos_list.append(f_pose_transformed)
else:
# select the depth located in (self.clothes_depth_arr_position)(ex 2/3), because some obstacle can cover person's clothes in front of person.
depth_list = []
for i in range(0, len(depth_arr)):
for j in range(0, len(depth_arr[0])):
if depth_arr[i][j] != 0:
depth_list.append(depth_arr[i][j])
if len(depth_list) == 0:
continue
sorted_depth_list = sorted(depth_list, reverse=False)
selected_idx = int(
len(sorted_depth_list) * self.clothes_depth_arr_position)
selected_depth = sorted_depth_list[selected_idx]
# personPose : [-y, x, z] by camera_link frame unit : mm
# data_to_send.data = rs2.rs2_deproject_pixel_to_point(self.intrinsics, [col, row], selected_depth) # [col, row]
rs2_pose = rs2.rs2_deproject_pixel_to_point(
self.intrinsics, [col, row], selected_depth) # [col, row]
pose_stamped = PoseStamped()
pose_stamped.header.frame_id = 'camera_link'
# personPose : [-y, x, z] unit : mm
pose_stamped.pose.position.x = rs2_pose[1] / 1000
pose_stamped.pose.position.y = -rs2_pose[0] / 1000
pose_stamped.pose.position.z = rs2_pose[2] / 1000
pose_stamped.pose.orientation.z = 0.0
pose_stamped.pose.orientation.w = 1.0
pose_stamped.header.stamp = rospy.Time.now()
pose_transformed = tf2_geometry_msgs.do_transform_pose(
pose_stamped, camera_transform)
pos_list.append(pose_transformed)
print('--------------------')
for idx, val in enumerate(self.person_clothes_hists):
name = self.person_names[idx]
if name.split('_')[0] == person_name:
# HISTCMP_INTERSECT = 2
similarity = cv2.compareHist(hist, val, 2)
if similarity > self.clothes_similarity:
found_target = True
#print(name)
#print(similarity)
#print('col : ', col, 'row : ', row)
#print(selected_depth)
#print('data_to_send : ', data_to_send)
send_pose_transformed = None
if use_face_area:
send_pose_transformed = f_pose_transformed
else:
send_pose_transformed = pose_transformed
#print('pose_transformed : ', send_pose_transformed)
self.personPose_pub.publish(send_pose_transformed)
self.last_person_pose = [send_pose_transformed, time_]
self.unauth_case_cnt = self.init_unauth_case_num
break
if found_target:
break
# when we can't find target by clothes histogram, then we try to check all person's position.
# if specific position is near from last target position, then, we can use this as new target position.
if (not found_target) and (self.last_person_pose is not None) and (
self.unauth_case_cnt != 0):
last_time = self.last_person_pose[1]
time_interval = time_ - last_time
if time_interval.to_sec() < self.valid_time_from_last_person:
if len(pos_list) != 0:
last_x = self.last_person_pose[0].pose.position.x
last_y = self.last_person_pose[0].pose.position.y
min_vel = None
min_idx = None
for idx, pos in enumerate(pos_list):
pos_x = pos.pose.position.x
pos_y = pos.pose.position.y
inc_x = pos_x - last_x
inc_y = pos_y - last_y
euclidean_distance = sqrt(
pow((inc_x), 2) + pow((inc_y), 2))
velocity = euclidean_distance / time_interval.to_sec()
if min_vel is None:
min_vel = velocity
min_idx = idx
elif velocity < min_vel:
min_vel = velocity
min_idx = idx
if min_vel < self.valid_vel:
self.personPose_pub.publish(pos_list[min_idx])
self.last_person_pose = [pos_list[min_idx], time_]
self.unauth_case_cnt -= 1
self.frame = frame
#Chi-square distance is one of the distance measures that can be used as a measure of dissimilarity between two histograms
#maxima distancia adotada
maxDist = 200
while True:
#return_value = 0 for some error or 1 for successful reading
#image is the frame capturedx
return_value, image = camera.read()
if return_value:
gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
cv2.imshow("image", gray_image)
H1 = cv2.calcHist( gray_image, [0], None, [256], [0,256])
H2 = cv2.calcHist(gray_image, [0], None, [256], [0,256])
comp = cv2.compareHist(H1, H2, cv2.HISTCMP_CHISQR)
height, width = gray_image.shape
while comp >= maxDist:
for x in range(height):
gray_image[x,0] = 255
gray_image[x, width-1] = 255
for y in range(width):
gray_image[0,y] = 255
gray_image[height-1, y] = 255
cv2.imshow("image", gray_image)
H1 = cv2.calcHist( gray_image, [0], None, [256], [0,256])
H2 = cv2.calcHist(gray_image, [0], None, [256], [0,256])
allFaces.append(
Face(f_hist, f_hsv, (x, y - 10, w, h + 10),
frame[y - 10:y + h, x:x + w]))
# ========================================================================
if len(speakers) == 0:
for face in allFaces:
roll += 1
face.identity = roll
speakers.append(Person(face.f_image, face.f_window, roll))
else:
for face in allFaces:
d = 0
coeff = 1
for speaker in speakers:
d = cv2.compareHist(face.f_hist, speaker.hist,
cv2.HISTCMP_BHATTACHARYYA)
if coeff > d and d < 0.4:
coeff = d
face.identity = speaker.roll
i.append(speaker.roll)
for speaker in speakers:
if speaker.roll == face.identity:
speaker.image = face.f_image
speaker.hist = face.f_hist
speaker.hsv = face.f_hsv
speaker.active = True
# if still no face has matched
if face.identity == -1:
roll += 1
def processQueryImage(self):
paramList = list()
with open(self.paramTxt) as f:
for line in f:
paramList.append(int(line.strip()))
print(paramList)
for queryImage in self.queryDict.iterkeys():
img = cv2.imread(queryImage)
imgGray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# radius = 3
# noPoints = 8 * radius
radius = paramList[0]
noPoints = paramList[1] * radius
lbp = local_binary_pattern(imgGray,
noPoints,
radius,
method='uniform')
# Calculate the histogram
x = itemfreq(lbp.ravel())
# normalize the histogram
queryHist = x[:, 1] / sum(x[:, 1])
results = []
for index, trainedHist in enumerate(self.lbpHistogram):
# Distance Metric to be used, these two Low score means good
score = cv2.compareHist(
np.array(trainedHist, dtype=np.float32),
np.array(queryHist, dtype=np.float32), cv2.HISTCMP_CHISQR)
results.append((self.addrImg[index], round(score, 3)))
#data = [('abc', 121),('qwe', 231),('pop', 148), ('gfh',221)]
#sorted(data, key=lambda x:x[1])
#sorted(data, key=lambda x:x[1], reverse=True)
results = sorted(results, key=lambda score: score[1])
self.results_all[queryImage] = results
print("Displaying scores for {} ** \n".format(queryImage))
for k in self.materialCode.keys():
if k in queryImage:
# print(k, end=" ")
self.trueList.append(self.materialCode.get(k))
for k in self.materialCode.keys():
if k in results[0][0]:
# print(k, end=" ")
self.predList.append(self.materialCode.get(k))
for image, score in results:
print("{} has score {}".format(image, score))
# font = cv2.FONT_HERSHEY_SIMPLEX
# cv2.putText(img, 'Bitumin', (10, 450), font, 1, (255, 255, 255), 2)
# cv2.putText(img, 'Gravel', (10, 450), font, 1, (255, 255, 255), 2, cv2.LINE_AA)
#################################################### Code added to show in matplot lib
# plt.axis("off")
# plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
# # plt.imshow(img)
# plt.show()
######################################################
cv2.imshow("query", img)
################
# print(results)
# print("SASAASAS" + results[0][0])
# test = cv2.imread(results[0][0])
# if not test is None:
# w, h = test.shape[:2]
# print(w)
# print(h)
# cv2.imshow(str(results[0][1]), cv2.imread(results[0][0]))
# cv2.waitKey()
# cv2.destroyAllWindows()
# else:
# print("Not read")
################
tmpImg = cv2.imread(results[0][0])
font = cv2.FONT_HERSHEY_SIMPLEX
cv2.putText(tmpImg, 'Bitumin', (10, 450), font, 1, (255, 255, 255),
2)
# plt.axis("off")
# plt.imshow(cv2.cvtColor(tmpImg, cv2.COLOR_BGR2RGB))
# plt.imshow(tmpImg)
# plt.show()
cv2.imshow(str(results[0][1]), tmpImg)
cv2.waitKey()
cv2.destroyAllWindows()
# ref.append(all_images[i_mod][0])
# wm_masks_ref.append(all_labels[0][0]) # although it's the same for all timepoints
#curr_ref = ref[i_mod] #Always first timepoint of first patient as reference
curr_ref = all_ref[i_mod]
curr_ref_mask = all_ref_labels[0]
#curr_ref_mask = wm_masks_ref[i_mod]
#save_image(curr_ref, jp(path_write, pat, modalities[i_mod] + "_norm_" + str(0+1).zfill(2)))
#histograms_aligned[i_mod][0] = histograms[i_mod][0] # First timepoint as reference(does not change)
for i_tp in range(
0, num_tp): # for each timepoint starting at the second one
curr_img = all_images[i_mod][i_tp]
curr_wm_mask = all_labels[0][i_tp]
f = lambda x: cv2.compareHist(
np.histogram(curr_ref[curr_ref_mask > 0].ravel(),
256, [0, 1],
density=True)[0].astype(np.float32),
np.histogram((x[0] * curr_img[curr_wm_mask > 0.5]).ravel(),
256, [0, 1],
density=True)[0].astype(np.float32), 1)
# Optimize Chi-Square metric
xopt = fmin(func=f, x0=[0.5])
curr_img_new = np.clip(xopt[0] * curr_img, 0, 1)
save_image(
curr_img_new,
jp(
path_write, pat, modalities[i_mod] + "_norm_" +
str(i_tp + 1).zfill(2) + ".nii.gz"))
all_images_aligned[i_mod][i_tp] = curr_img_new
histograms_aligned[i_mod][i_tp] = np.histogram(
curr_img_new[curr_img_new > 0].ravel(),
256, [0, 1],
#! /usr/bin/python3 -u # Returns the similarity of 2 images. import cv2, sys i1 = cv2.imread(sys.argv[1], 0) i2 = cv2.imread(sys.argv[2], 0) channels = [0] histSize = [256] mask = None ranges = [0, 256] h1 = cv2.calcHist([i1], channels, mask, histSize, ranges) h2 = cv2.calcHist([i2], channels, mask, histSize, ranges) method = cv2.HISTCMP_CORREL print(cv2.compareHist(h1, h2, method))
img = cv2.imread(os.path.join(test_path, file),
cv2.IMREAD_GRAYSCALE)
height = img.shape[1]
if USE_STEP:
step = int(nameValue[5])
origin = img[y + height:y + step + height, x:x + step]
imgOut = img[y:y + step, x:x + step]
else:
origin = img[height:, :]
imgOut = img[:height, :]
originHist = cv2.calcHist([origin], [0], None, [hist_size],
[0.0, 255.0])
imgHist = cv2.calcHist([imgOut], [0], None, [hist_size],
[0.0, 255.0])
dist1 = cv2.compareHist(originHist, imgHist, cv2.HISTCMP_CORREL)
dist2 = cv2.compareHist(originHist, imgHist, cv2.HISTCMP_CHISQR)
dist3 = cv2.compareHist(originHist, imgHist, cv2.HISTCMP_INTERSECT)
dist4 = cv2.compareHist(originHist, imgHist,
cv2.HISTCMP_BHATTACHARYYA)
dist5 = cv2.compareHist(originHist, imgHist, cv2.HISTCMP_HELLINGER)
psnr = skimage.measure.compare_psnr(origin, imgOut)
ssim = skimage.measure.compare_ssim(origin, imgOut)
fid = calFID(origin, imgOut, sess)
mutual = mr.mutual_info_score(np.reshape(origin, -1),
np.reshape(imgOut, -1))
outFile.write(file + ',' + str(dist1) + ',' + str(dist2) + ',' +
str(dist3) + ',' + str(dist4) + ',' + str(dist5) +
',' + str(ssim) + ',' + str(psnr) + ',' + str(fid) +
',' + str(mutual) + '\n')
def histogram_comparator(X, Y):
n_rows = Y.shape[1]
n_cols = Y.shape[2]
row_wnd = int(np.floor(X.shape[0] / n_rows))
col_wnd = int(np.floor(X.shape[1] / n_cols))
mse = []
c = 0
# X = calculate_spatial_histogram(X, n_rows, n_cols, CONFIG['n_hist_bins'])
# idxs = np.where(X > 0)
# print(X.shape, Y.shape)
# diff = Y - X
# diff = np.moveaxis(diff, 0, len(diff.shape) - 1)
#
# diff = diff[idxs]
#
# # Manhattan
# # res = np.sum(np.abs(res), axis=(0,1,2,3,4))
#
# print(diff.shape)
#
# res = -np.sum(diff**2 / diff, axis=(0))
#
# X = X.reshape()
# res = np.reshape(res, (CONFIG['n_hist_bins']**3 * n_rows * n_cols, res.shape[-1:][0]))
# mse = res
# BHATTACHARYYA
for x in range(n_rows):
for y in range(n_cols):
h = calculate_histogram(X[
x * row_wnd: (x + 1) * row_wnd,
y * col_wnd: (y + 1) * col_wnd
], n_bins=CONFIG['n_hist_bins']).astype(np.float16)
r = []
if np.sum(h) > 0:
if X.shape[2] == 3:
for i in range(Y.shape[0]):
a = h
b = Y[i, x, y]
d = cv2.compareHist(a.astype(np.float32), b.astype(np.float32), cv2.HISTCMP_BHATTACHARYYA)
r.append(d)
else:
idxs = np.where(h > 0)
for i in range(Y.shape[0]):
a = h[idxs]
b = Y[i, x, y][idxs]
d = cv2.compareHist(a.astype(np.float32), b.astype(np.float32), cv2.HISTCMP_BHATTACHARYYA)
r.append(d)
c += 1
else:
# r = [np.inf] * Y.shape[0]
r = [0] * Y.shape[0]
mse.append(r)
# mse = np.array(mse)
# mean = np.mean(mse[np.where(mse!=np.inf)])
#
# for i in range(n_rows * n_cols):
# if np.any(mse[i] == np.inf):
# print("replacing")
# mse[i, :] = mean
# mse = np.mean(mse, axis=0)
mse = np.sum(np.array(mse), axis=0)
return mse
sample['avc3'] = float(average_color[2])
# area / rect
sample['extent'] = sample['m00'] / (h * w)
#min_rect / rect
(min_x, min_y), (min_w, min_h), min_theta = cv2.minAreaRect(c)
sample['rect_extent'] = (min_w * min_h) / (h * w)
#print(sample['m00'], min_w*min_h, h*w)
#sys.exit(0)
hist = cv2.calcHist([crop_img], [0], None, [256], [0, 256])
sample['hist_correl'] = cv2.compareHist(
hist / np.linalg.norm(hist), average_normed_hist,
cv2.HISTCMP_CORREL)
sample['hist_chisqr'] = cv2.compareHist(
hist / np.linalg.norm(hist), average_normed_hist,
cv2.HISTCMP_CHISQR)
sample['hist_bhatt'] = cv2.compareHist(hist / np.linalg.norm(hist),
average_normed_hist,
cv2.HISTCMP_BHATTACHARYYA)
sample['hist_inter'] = cv2.compareHist(hist / np.linalg.norm(hist),
average_normed_hist,
cv2.HISTCMP_INTERSECT)
# 'white_balance'
#print(int(hist[-1])/sample['m00'])
sample['white_balance'] = int(hist[-1]) / (h * w)
def calc_hist_score(hist1, hist2):
scores = []
for channel1, channel2 in zip(hist1, hist2):
score = cv2.compareHist(channel1, channel2, cv2.HISTCMP_BHATTACHARYYA)
scores.append(score)
return np.mean(scores)
def main():
path = 'paintings/'
orb = cv2.ORB(1000, 1.2)
#dictionary for holding the histograms
index = {}
#dictionary for holding the RGB images
images = {}
#dictionary for holding the OBJ descriptors
bagdex = {}
#dictionary for holding the OBJ keypoints
bagdex_kp = {}
#dictionary for holding greyscale images
greyges = {}
for filename in os.listdir(path):
if filename.endswith('.jpg'):
print filename
image = cv2.imread(path + filename)
#histogram data
images[filename] = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
hist = cv2.calcHist([image], [0, 1, 2], None, [8, 8, 8],
[0, 256, 0, 256, 0, 256])
hist = cv2.normalize(hist).flatten()
index[filename] = hist
#OBJ feature descriptors data
image = cv2.imread(path + filename, cv2.CV_LOAD_IMAGE_GRAYSCALE)
greyges[filename] = image
orb = cv2.ORB(250, 1.2)
kp, des = orb.detectAndCompute(greyges[filename], None)
bagdex[filename] = des
bagdex_kp[filename] = kp
active = True
while (active):
query_filepath = raw_input("Please enter filepath of query image: \n")
ind = query_filepath.find('/')
query_filename = query_filepath
if ind != -1:
query_filename = query_filepath[ind + 1:]
# Create histogram for query image
query_image = cv2.imread(query_filepath)
images[query_filename] = cv2.cvtColor(query_image, cv2.COLOR_BGR2RGB)
query_hist = cv2.calcHist([query_image], [0, 1, 2], None, [8, 8, 8],
[0, 256, 0, 256, 0, 256])
query_hist = cv2.normalize(query_hist).flatten()
index[query_filename] = query_hist
results = {}
# Create ORB feature descriptor for query image
query_bagage = cv2.imread(query_filepath, cv2.CV_LOAD_IMAGE_GRAYSCALE)
greyges[query_filename] = query_bagage
query_kp, query_des = orb.detectAndCompute(greyges[query_filename],
None)
bagdex[query_filename] = query_des
bagdex_kp[query_filename] = query_kp
bagsults = {}
# initialize the comparison method for histogram
methodName = "Hellinger"
method = cv2.cv.CV_COMP_BHATTACHARYYA
for (k, hist) in index.items():
# compute the distance between the two histograms
# using the method and update the results dictionary
d = cv2.compareHist(index[query_filename], hist, method)
results[k] = d
for (k, des) in bagdex.items():
d = 0
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(bagdex[query_filename], des)
matches = sorted(matches, key=lambda val: val.distance)
for match in matches[:10]:
d += match.distance
bagsults[k] = d
# normalize the results
hist_sum = 0
bag_sum = 0
for (name, value) in results.items():
hist_sum += results[name]
bag_sum += bagsults[name]
for (name, value) in results.items():
results[name] = value / float(hist_sum)
bagsults[name] = bagsults[name] / float(bag_sum)
total_results = {}
for (name, value) in results.items():
total_results[name] = value + bagsults[name]
# sort the results
total_results = sorted([(v, k) for (k, v) in total_results.items()],
reverse=False)
results = sorted([(v, k) for (k, v) in results.items()], reverse=False)
bagsults = sorted([(v, k) for (k, v) in bagsults.items()],
reverse=False)
# initialize the results figure
fig = plt.figure("Results: ")
# loop over the results
n = 0
for (i, (v, k)) in enumerate(bagsults):
if n > 9:
break
n += 1
# show the result
ax = fig.add_subplot(1, 10, i + 1)
ax.set_title("%s: %.2f" % (k, v))
plt.imshow(images[k])
plt.axis("off")
# show the matching lines drawn instead
#out = drawMatches(query_bagage, query_kp, greyges[k], bagdex_kp[k], matches[:10])
#downloads the image into our own database
#copyfile(query_filepath, 'paintings/copy_' + query_filename)
# show the results
plt.show()
def checkliver():# 사용자 이미지 검사
print(user_file)#사용자 파일 경로 가져옴
p_var2.set(5)
progress_bar2.update()
image = cv2.imread(user_file)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
image = image.reshape((image.shape[0] * image.shape[1], 3))
k = 30 # 색상 K개 추출
clt = KMeans(n_clusters=k)
clt.fit(image)
hist = centroid_histogram(clt)
colorSort = clt.cluster_centers_
bar = plot_colors(hist, colorSort)
global photo_userimgcheck # 사용자 check이미지를 사용하기 위해서 global로 해주어야 함
usercheckimg = Image.fromarray(bar)#넘파이를 이미지로 바꿔줌
photo_userimgcheck = ImageTk.PhotoImage(image=usercheckimg) # bar을 pilimage로 바꿔주고 tk로 열어주어야함
userimg_bar.config(image=photo_userimgcheck,width=580,height=70)
# #이미지 분석하는 코드
img_original = bar#원본
img1 = cv2.imread(checkliver1)#1단계
img2 = cv2.imread(checkliver2)#2단계
img3 = cv2.imread(checkliver3)#3단계
img4 = cv2.imread(checkliver4)#4단계
p_var2.set(90)
progress_bar2.update()
#원본 정규화
img_original = cv2.cvtColor(img_original, cv2.COLOR_BGR2HSV)
img_original = cv2.calcHist([img_original], [0, 1], None, [180, 256], [0, 180, 0, 256])
img_original = cv2.normalize(img_original, None, 0, 1, cv2.NORM_MINMAX)
imgs = [img1, img2, img3, img4]# 단계 사진만 구분해서 정규화함
hists = []
for i, img in enumerate(imgs):
#1 각 이미지를 HSV로 변환
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
#2 H,S 채널에 대한 히스토그램 계산
hist = cv2.calcHist([hsv], [0, 1], None, [180, 256], [0, 180, 0, 256])
#3 0~1로 정규화
cv2.normalize(hist, hist, 0, 1, cv2.NORM_MINMAX)
hists.append(hist)
#query = hists[0]#원본 이미지
correl_result = []
chisqr_result = []
intersect_result = []
bhattacharyya_result = []
correl = cv2.HISTCMP_CORREL
chisqr = cv2.HISTCMP_CHISQR
intersect = cv2.HISTCMP_INTERSECT # ret = ret / np.sum(query) , 교차 분석인 경우 비교대상으로 나누어 1로 정규화
bhattacharyya = cv2.HISTCMP_BHATTACHARYYA
listbox.insert(0, '선택한 이미지' + filenm[1])
#코렐 분석
for hist in hists:
c_rs = cv2.compareHist(img_original, hist, correl)
correl_result.append(c_rs)
correl_result = np.array(correl_result)
correl_similar = find_nearest(correl_result, 1.0)#1.0과 가장 가까운 값 찾음
correl_result = correl_result.tolist()
correl_similar_index = correl_result.index(correl_similar)#가장 비슷한 이미지 위치 반환
#print('해당 이미지는',correl_similar_index+1,'단계 입니다.')
listbox.insert(1, '코렐분석 결과 해당 이미지는' + str(correl_similar_index+1)+' 단계 입니다.')
#카이제곱
for hist in hists:
ch_rs = cv2.compareHist(img_original, hist, chisqr)
chisqr_result.append(ch_rs)
chisqr_result = np.array(chisqr_result)
chisqr_similar = find_nearest(chisqr_result, 0)#0과 가장 가까운 값 찾음
chisqr_result = chisqr_result.tolist()
chisqr_similar_index = chisqr_result.index(chisqr_similar)#가장 비슷한 이미지 위치 반환
#print('해당 이미지는',correl_similar_index+1,'단계 입니다.')
listbox.insert(2, '카이제곱분석 결과 해당 이미지는' + str(chisqr_similar_index+1)+' 단계 입니다.')
#교차검증
for hist in hists:
intersect_rs = cv2.compareHist(img_original, hist, intersect)
intersect_result.append(intersect_rs/np.sum(img_original))
intersect_result = np.array(intersect_result)
intersect_similar = find_nearest(intersect_result, 1.0)#1.0과 가장 가까운 값 찾음
intersect_result = intersect_result.tolist()
intersect_similar_index = intersect_result.index(intersect_similar)#가장 비슷한 이미지 위치 반환
#print('해당 이미지는',correl_similar_index+1,'단계 입니다.')
listbox.insert(3, '교차검증 결과 해당 이미지는' + str(intersect_similar_index+1)+' 단계 입니다.')
#바타차야 거리
for hist in hists:
bhattacharyya_rs = cv2.compareHist(img_original, hist, bhattacharyya)
bhattacharyya_result.append(bhattacharyya_rs)
bhattacharyya_result = np.array(bhattacharyya_result)
bhattacharyya_similar = find_nearest(bhattacharyya_result, 0)#0과 가장 가까운 값 찾음
bhattacharyya_result = bhattacharyya_result.tolist()
bhattacharyya_similar_index = bhattacharyya_result.index(bhattacharyya_similar)#가장 비슷한 이미지 위치 반환
#print('해당 이미지는',correl_similar_index+1,'단계 입니다.')
listbox.insert(4, '바타차야 거리 분석 결과 해당 이미지는' + str(bhattacharyya_similar_index+1)+' 단계 입니다.')
#총 갯수 출력해주기
total_list = [correl_similar_index,chisqr_similar_index,intersect_similar_index,bhattacharyya_similar_index]
total_list = np.array(total_list)
total_list_count = np.bincount(total_list)#토탈 값 카운팅
print(total_list_count)
total_list_max = np.max(total_list_count)# 카운팅 중 가장 높은 수
print(total_list_max)
total_list_count = total_list_count.tolist()
total_list_index = total_list_count.index(total_list_max)
print(total_list_index)
if total_list_max == 2:
print('정확한 측정이 어렵습니다. 분석 이미지를 바꿔주세요')
listbox.insert(5, '정확한 측정이 어렵습니다. 분석 이미지를 바꿔주세요')
listbox.insert(6, '======================================================================================')
else:
print('결과는',(total_list_index+1),'입니다')
listbox.insert(5, '해당 이미지는 ' + str(total_list_index+1) + ' 단계 입니다.')
listbox.insert(6, '======================================================================================')
#표시하는 부분
userimg = Label(userimg_frame, image=logo)
p_var2.set(100)
progress_bar2.update()
def get_avg_score_hist(img1, references):
scores = []
for img2 in references:
score = cv2.compareHist(img1, img2, cv2.HISTCMP_BHATTACHARYYA)
scores.append(score)
return sum(scores) / len(scores)
def check_position(self,
hist_thresh=(75, 80),
pos_thresh=(10, 15),
metric=cv2.TM_CCOEFF_NORMED,
display=False,
test=False,
roi=None,
pct_thresh=True):
"""Check camera is positioned correctly
For the template matching zero-normalized cross-correlation (default) should be more
robust to exposure (which we're not checking here). The L2 norm (TM_SQDIFF) should
also work.
If display is True, the template ROI (pick hashed) is plotted over a video frame,
along with the threshold regions (green solid). The histogram correlations are plotted
and the full histogram is plotted for one of the sample frames and the reference frame.
:param hist_thresh: The minimum histogram cross-correlation threshold to pass (0-1).
:param pos_thresh: The maximum number of pixels off that the template matcher may be off
by. If two values are provided, the lower threshold is treated as a warning boundary.
:param metric: The metric to use for template matching.
:param display: If true, the results are plotted
:param test: If true a reference frame instead of the frames in frame_samples.
:param roi: A tuple of indices for the face template in the for ((y1, y2), (x1, x2))
:param pct_thresh: If true, the thresholds are treated as percentages
"""
if not test and self.data['frame_samples'] is None:
return 'NOT_SET'
refs = self.load_reference_frames(self.side)
# ensure iterable
pos_thresh = np.sort(np.array(pos_thresh))
hist_thresh = np.sort(np.array(hist_thresh))
# Method 1: compareHist
ref_h = cv2.calcHist([refs[0]], [0], None, [256], [0, 256])
frames = refs if test else self.data['frame_samples']
hists = [cv2.calcHist([x], [0], None, [256], [0, 256]) for x in frames]
corr = np.array([
cv2.compareHist(test_h, ref_h, cv2.HISTCMP_CORREL)
for test_h in hists
])
if pct_thresh:
corr *= 100
hist_passed = [np.all(corr > x) for x in hist_thresh]
# Method 2:
top_left, roi, template = self.find_face(roi=roi,
test=test,
metric=metric,
refs=refs)
(y1, y2), (x1, x2) = roi
err = (x1, y1) - np.median(np.array(top_left), axis=0)
h, w = frames[0].shape[:2]
if pct_thresh: # Threshold as percent
# t_x, t_y = pct_thresh
err_pct = [(abs(x) / y) * 100 for x, y in zip(err, (h, w))]
face_passed = [all(err_pct < x) for x in pos_thresh]
else:
face_passed = [np.all(np.abs(err) < x) for x in pos_thresh]
if display:
plt.figure()
# Plot frame with template overlay
img = frames[0]
ax0 = plt.subplot(221)
ax0.imshow(img, cmap='gray', vmin=0, vmax=255)
bounds = (x1 - err[0], x2 - err[0], y2 - err[1], y1 - err[1])
ax0.imshow(template, cmap='gray', alpha=0.5, extent=bounds)
if pct_thresh:
for c, thresh in zip(('green', 'yellow'), pos_thresh):
t_y = (h / 100) * thresh
t_x = (w / 100) * thresh
xy = (x1 - t_x, y1 - t_y)
ax0.add_patch(
Rectangle(xy,
x2 - x1 + (t_x * 2),
y2 - y1 + (t_y * 2),
fill=True,
facecolor=c,
lw=0,
alpha=0.05))
else:
for c, thresh in zip(('green', 'yellow'), pos_thresh):
xy = (x1 - thresh, y1 - thresh)
ax0.add_patch(
Rectangle(xy,
x2 - x1 + (thresh * 2),
y2 - y1 + (thresh * 2),
fill=True,
facecolor=c,
lw=0,
alpha=0.05))
xy = (x1 - err[0], y1 - err[1])
ax0.add_patch(
Rectangle(xy,
x2 - x1,
y2 - y1,
edgecolor='pink',
fill=False,
hatch='//',
lw=1))
ax0.set(xlim=(0, img.shape[1]), ylim=(img.shape[0], 0))
ax0.set_axis_off()
# Plot the image histograms
ax1 = plt.subplot(212)
ax1.plot(ref_h[5:-1], label='reference frame')
ax1.plot(np.array(hists).mean(axis=0)[5:-1], label='mean frame')
ax1.set_xlim([0, 256])
plt.legend()
# Plot the correlations for each sample frame
ax2 = plt.subplot(222)
ax2.plot(corr, label='hist correlation')
ax2.axhline(hist_thresh[0],
0,
self.n_samples,
linestyle=':',
color='r',
label='fail threshold')
ax2.axhline(hist_thresh[1],
0,
self.n_samples,
linestyle=':',
color='g',
label='pass threshold')
ax2.set(xlabel='Sample Frame #', ylabel='Hist correlation')
plt.legend()
plt.suptitle('Check position')
plt.show()
pass_map = {i: s for i, s in enumerate(('FAIL', 'WARNING', 'PASS'))}
face_aligned = pass_map[sum(face_passed)]
hist_correlates = pass_map[sum(hist_passed)]
return self.overall_outcome([face_aligned, hist_correlates])
) # create gray scale histogram, removing black color
tot_pixel = histWB.sum() # calculate not black pixel
histDetection_perc[0] = (histBlue / tot_pixel
) # normalize hist
histDetection_perc[1] = (histGreen / tot_pixel
) # normalize hist
histDetection_perc[2] = (histRed / tot_pixel
) # normalize hist
if (num_detection_squad_1 != 0):
hist_squad_1_perc = normalizeHist(
hist_squad_1, num_detection_squad_1
) # normalize and mediates hist
compHist_1 = cv2.compareHist(
np.float32(hist_squad_1_perc),
np.float32(histDetection_perc),
cv2.HISTCMP_BHATTACHARYYA) # compare hist
if (num_detection_squad_2 != 0):
hist_squad_2_perc = normalizeHist(
hist_squad_2, num_detection_squad_2
) # normalize and mediates hist
compHist_2 = cv2.compareHist(
np.float32(hist_squad_2_perc),
np.float32(histDetection_perc),
cv2.HISTCMP_BHATTACHARYYA) # compoare hist
if (num_detection_referee != 0):
hist_referee_perc = normalizeHist(
hist_referee, num_detection_referee
) # normalize and mediates hist
compHist_3 = cv2.compareHist(
def spaciograms_distance_rating(spaciogram_1, spaciogram_2, rank):
'''
:param spaciogram_1:
:param spaciogram_2:
:param rank:
:return:
'''
############ CHECKS ############
# check if spaciogram_1.shape == spaciogram_2.shape:
rating = []
# spaciogram_1 = np.array(spaciogram_1)
# spaciogram_2 = np.array(spaciogram_2)
# if spaciogram_1.shape != spaciogram_2.shape is False:
# print 'Error: the dimensions of spaciogram_1 and spaciogram_2 are not equal! \n' \
# 'shapes are: 1st - ' + str(spaciogram_1.shape) + '\n' \
# 'shapes are: 2nd - ' + str(spaciogram_2.shape)
# return rating
if rank < 1 or rank > 3:
print 'Error: only 3 ranks, rank = 1, 2 or 3!'
return rating
# # Define number of rows (overall bin count):
# numRows = spaciogram_1.size
# dims = len(spaciogram_1.shape)
# bins_per_dim = len(spaciogram_1)
# signature_1 = np.zeros([numRows, dims+1]) #cv2.CreateMat(numRows, dims, cv2.CV_32FC1)
# print signature_1.shape
# signature_2 = signature_1 #cv2.CreateMat(numRows, dims, cv2.CV_32FC1)
# sigrature_index = 0
# # fill signature_natures:
# # TODO: for production optimize this, use Numpy (reshape?)
# for d1 in range(0, bins_per_dim - 1):
# for d2 in range(0, bins_per_dim - 1):
# for d3 in range(0, bins_per_dim - 1):
# for d4 in range(0, bins_per_dim - 1):
# for d5 in range(0, bins_per_dim - 1):
# # signature 1:
# signature_1[sigrature_index, :] = [spaciogram_1[d1, d2, d3, d4, d5], d1, d2, d3, d4, d5]
# # bin_val = cv2.QueryHistValue_2D(spaciogram_1, d1, d2, d3, d4, d5)
# # cv.Set2D(signature_1, sigrature_index, 0, bin_val) #bin value
# # cv.Set2D(signature_1, sigrature_index, 1, d1) #coord1
# # cv.Set2D(signature_1, sigrature_index, 2, d2) #coord2
# # cv.Set2D(signature_1, sigrature_index, 3, d3) #coord3
# # cv.Set2D(signature_1, sigrature_index, 4, d4) #coord4
# # cv.Set2D(signature_1, sigrature_index, 5, d5) #coord5
# # signature 2:
# signature_2[sigrature_index, :] = [spaciogram_2[d1, d2, d3, d4, d5], d1, d2, d3, d4, d5]
# # bin_val2 = cv2.QueryHistValue_2D(spaciogram_2, d1, d2, d3, d4, d5)
# # cv.Set2D(signature_2, sigrature_index, 0, bin_val2) #bin value
# # cv.Set2D(signature_2, sigrature_index, 1, d1) #coord1
# # cv.Set2D(signature_2, sigrature_index, 2, d2) #coord2
# # cv.Set2D(signature_2, sigrature_index, 3, d3) #coord3
# # cv.Set2D(signature_2, sigrature_index, 4, d4) #coord4
# # cv.Set2D(signature_2, sigrature_index, 5, d5) #coord5
# sigrature_index += 1
# print spaciogram_1[d1, d2, d3, d4, d5]
# signature_1 = np.zeros([spaciogram_1.size / len(spaciogram_1), len(spaciogram_1)])
# sigrature_index = 0
# # print len(spaciogram_1)
# for dim in spaciogram_1:
# signature_1[:, sigrature_index] = dim.flatten()
# sigrature_index += 1
#
# signature_2 = np.zeros([spaciogram_2.size / len(spaciogram_1), len(spaciogram_2)])
# sigrature_index = 0
# for dim in spaciogram_2:
# signature_2[:, sigrature_index] = dim.flatten()
# sigrature_index += 1
# signature_1 = np.reshape(spaciogram_1, (spaciogram_1[0].size, len(spaciogram_1)))
# signature_2 = np.reshape(spaciogram_2, (spaciogram_2[0].size, len(spaciogram_2)))
method = cv2.HISTCMP_CHISQR
# HISTCMP_CORREL Correlation
# HISTCMP_CHISQR Chi-Square
# HISTCMP_INTERSECT Intersection
# HISTCMP_BHATTACHARYYA Bhattacharyya distance
# HISTCMP_HELLINGER Synonym for HISTCMP_BHATTACHARYYA
# HISTCMP_CHISQR_ALT
# HISTCMP_KL_DIV
if rank != 3:
rating = cv2.compareHist(spaciogram_1, spaciogram_2, method)
# elif rank == 2:
# rating = cv2.compareHist(np.array(spaciogram_1[1]).astype('float32'),
# np.array(spaciogram_2[1]).astype('float32'), method)
elif rank == 3:
rating = 0.0
for i in range(2, len(spaciogram_1)):
rating += cv2.compareHist(spaciogram_1[i], spaciogram_2[i], method)
else:
rating = []
# rating = emd(signature_1, signature_2)
return rating
def detect_video(self, yolo):
from PIL import Image, ImageFont, ImageDraw
#Start ROS node
pub, pub_flag = start_node()
accum_time = 0
curr_fps = 0
fps = "FPS: ??"
prev_time = timer()
while True:
if self.ret:
frame = self.frames[0]
depth_frame = self.frames[1]
if (type(frame) is int)or(type(depth_frame) is int):
print("!!!CAUTION!!! type of frame is int")
continue
image = Image.fromarray(frame)
image, bottle, person, right, left, bottom, top, right2, left2, bottom2, top2 = yolo.detect_image(image, pub)
result = np.asarray(image)
curr_time = timer()
exec_time = curr_time - prev_time
prev_time = curr_time
accum_time = accum_time + exec_time
curr_fps = curr_fps + 1
if accum_time > 1:
accum_time = accum_time - 1
fps = "FPS: " + str(curr_fps)
curr_fps = 0
cv2.putText(result, text=fps, org=(3, 15), fontFace=cv2.FONT_HERSHEY_SIMPLEX,
fontScale=0.50, color=(255, 0, 0), thickness=2)
cv2.imshow("result", result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
if (bottle==False) or (person==False):
continue
# ------------------------------Tracking-----------------------------------
# tracker_types = ['BOOSTING', 'MIL','KCF', 'TLD', 'MEDIANFLOW', 'GOTURN', 'MOSSE', 'CSRT']
# tracker_type = tracker_types[7]
tracker = cv2.TrackerCSRT_create()
tracker2 = cv2.TrackerCSRT_create()
# setup initial location of window
left, right, top, bottom = left, right, top, bottom
r,h,ci,w = top, bottom-top, left, right-left # simply hardcoded the values r, h, c, w
import matplotlib.pyplot as plt
# frame_hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
frame_b, frame_g, frame_r = frame[:,:,0], frame[:,:,1], frame[:,:,2]
hist_b = cv2.calcHist([frame_b[top:bottom, left:right]],[0],None,[256],[0,256])
hist_g = cv2.calcHist([frame_g[top:bottom, left:right]],[0],None,[256],[0,256])
hist_r = cv2.calcHist([frame_r[top:bottom, left:right]],[0],None,[256],[0,256])
cv2.normalize(hist_b, hist_b,0,255,cv2.NORM_MINMAX)
cv2.normalize(hist_g, hist_g,0,255,cv2.NORM_MINMAX)
cv2.normalize(hist_r, hist_r,0,255,cv2.NORM_MINMAX)
# plt.plot(hist_r, color='r', label="r")
# plt.plot(hist_g, color='g', label="g")
# plt.plot(hist_b, color='b', label="b")
# plt.show()
track_window = (ci, r, w, h)
r2,h2,ci2,w2 = top2, bottom2-top2, left2, right2-left2 # simply hardcoded the values r, h, c, w
track_window2 = (ci2, r2, w2, h2)
cv2.imwrite('bottledetect.jpg', frame[r:r+h, ci:ci+w])
cv2.imwrite('persondetect.jpg', frame[r2:r2+h2, ci2:ci2+w2])
# set up the ROI for tracking
roi = frame[r:r+h, ci:ci+w]
hsv_roi = cv2.cvtColor(roi, cv2.COLOR_BGR2HSV)
mask = cv2.inRange(hsv_roi, np.array((0., 60.,32.)), np.array((180.,255.,255.)))
roi_hist = cv2.calcHist([hsv_roi],[0],mask,[180],[0,180])
cv2.normalize(roi_hist,roi_hist,0,255,cv2.NORM_MINMAX)
# Setup the termination criteria, either 10 iteration or move by atleast 1 pt
term_crit = ( cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 1 )
ok = tracker.init(frame, track_window)
ok2 = tracker2.init(frame, track_window2)
track_thing = 0 #bottle
pts = Point()
pts2 = Point()
untrack = 0
while(1):
if self.ret:
frame = self.frames[0]
depth_frame = self.frames[1]
depth = depth_frame
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
dst = cv2.calcBackProject([hsv],[0],roi_hist,[0,180],1)
# apply meanshift to get the new location
print(track_window2)
ok, track_window = tracker.update(frame)
x,y,w,h = track_window
ok, track_window2 = tracker2.update(frame)
x2,y2,w2,h2 = track_window2
# Draw it on image
img2 = cv2.rectangle(frame, (x,y), (x+w,y+h), 255,2)
if not track_thing:
img2 = cv2.rectangle(img2, (x2,y2), (x2+w2,y2+h2), 255,2)
else:
img2 = cv2.rectangle(img2, (x2, y2), (x2+w2, y2+h2),(0, 0, 255), 2)
cv2.imshow('Tracking',img2)
# https://www.intelrealsense.com/wp-content/uploads/2020/06/Intel-RealSense-D400-Series-Datasheet-June-2020.pdf
total, cnt = 0, 0
for i in range(3):
for j in range(3):
# dep = depth[j+y+h//2, i+x+w//2]*0.001
dep = depth[np.minimum((j+y+h//2), 479), np.minimum((i+x+w//2), 639)]*0.001
if (dep)!=0:
total += dep
cnt += 1
if cnt!=0:
worldz = total/cnt
else:
worldz = 0
total2, cnt2 = 0, 0
for i in range(3):
for j in range(3):
dep2 = depth[np.minimum((j+y2+h2//2), 479), np.minimum((i+x2+w2//2), 639)]*0.001
if dep2!=0:
total2 += dep2
cnt2 += 1
if cnt2!=0:
worldz2 = total2/cnt2
else:
worldz2 = 0
print('worldz', worldz)
print('worldz2', worldz2)
if (worldz == 0) or (worldz2 == 0):
worldx, worldy = 0, 0
pts.x, pts.y, pts.z = 0.0, 0.0, 0.0
worldx2, worldy2 = 0, 0
pts2.x, pts2.y, pts2.z = 0.0, 0.0, 0.0
else:
# focus length = 1.93mm, distance between depth cameras = about 5cm, a pixel size = 3um
if (track_thing==0):
#human Tracking
u_ud = (0.05*1.88*10**(-3))/(3*10**(-6)*worldz)
print('u_ud', u_ud)
# 深度カメラとカラーカメラの物理的な距離を考慮した項(-0.3*u_ud)
# これらの座標は物体を見たときの左の深度カメラを基準とする
worldx = 0.05*(x+w//2 - (img2.shape[1]//2) - 0.3*u_ud)/u_ud
worldy = 0.05*((img2.shape[0]//2) - (y+h))/u_ud
print('x,y,z = ', worldx, worldy, worldz)
pts.y, pts.z, pts.x = float(worldx), float(worldy), float(worldz)
else:
#bottle Tracking
u_ud = (0.05*1.88*10**(-3))/(3*10**(-6)*worldz2)
print('u_ud', u_ud)
worldx2 = 0.05*(x2+w2//2 - (img2.shape[1]//2) - 0.3*u_ud)/u_ud
worldy2 = 0.05*((img2.shape[0]//2) - (y2+h2))/u_ud
print('x2,y2,z2 = ', worldx2, worldy2, worldz2)
pts2.y, pts2.z, pts.x = float(worldx2), float(worldy2), float(worldz2)
print("track_thing = ", track_thing)
frame_b, frame_g, frame_r = frame[:,:,0], frame[:,:,1], frame[:,:,2]
hist_b2 = cv2.calcHist([frame_b[y: y+h, x: x+w]],[0],None,[256],[0,256])
hist_g2 = cv2.calcHist([frame_g[y: y+h, x: x+w]],[0],None,[256],[0,256])
hist_r2 = cv2.calcHist([frame_r[y: y+h, x: x+w]],[0],None,[256],[0,256])
# plt.plot(hist_r2, color='r', label="r")
# plt.plot(hist_g2, color='g', label="g")
# plt.plot(hist_b2, color='b', label="b")
# plt.show()
# if (track_window==(0, 0, 0, 0)) or (track_window2==(0, 0, 0, 0)):
cv2.normalize(hist_b2, hist_b2,0,255,cv2.NORM_MINMAX)
cv2.normalize(hist_g2, hist_g2,0,255,cv2.NORM_MINMAX)
cv2.normalize(hist_r2, hist_r2,0,255,cv2.NORM_MINMAX)
print('compareHist(b)', cv2.compareHist(hist_b, hist_b2, cv2.HISTCMP_CORREL))
print('compareHist(g)', cv2.compareHist(hist_g, hist_g2, cv2.HISTCMP_CORREL))
print('compareHist(r)', cv2.compareHist(hist_r, hist_r2, cv2.HISTCMP_CORREL))
print('track_window = ', track_window)
if ((cv2.compareHist(hist_b, hist_b2, cv2.HISTCMP_CORREL)<=0.3)or(cv2.compareHist(hist_g, hist_g2, cv2.HISTCMP_CORREL)<=0.3)or(cv2.compareHist(hist_r, hist_r2, cv2.HISTCMP_CORREL)<=0.3))or((track_window==(0, 0, 0, 0)) or (track_window2==(0, 0, 0, 0))):
untrack += 1
print("untrack = ", untrack)
if untrack>=30:
print("追跡が外れた！\n")
break
if ((worldy<=-0.5) and (not track_thing)):
print("ポイ捨てした！\n")
track_thing = 1 #human
if track_thing==0:
tracking_point = pts
flag = 0 #bottle
else:
tracking_point = pts2
flag = 1 #person
pub.publish(tracking_point)
pub_flag.publish(flag)
k = cv2.waitKey(60) & 0xff
if k == 27:
break
else:
break
yolo.close_session()
def score_images(self, im1, im2, method="orb"):
"""Score the similarity between two images according to differents methods.
im1 = framed photo ; im2 = db_im"""
score = 0
if method == "ssim":
im1 = cv2.cvtColor(im1, cv2.COLOR_BGR2GRAY)
im2 = cv2.cvtColor(im2, cv2.COLOR_BGR2GRAY)
score = ssim(cv2.resize(im1, (im2.shape[1], im2.shape[0]), interpolation = cv2.INTER_AREA), im2, multichannel=False)
if method == "hist_inter":
#crop_zone = (20,35,203,171)
#im1 = im1[20:203, 35:171]
#im2 = im2[20:203, 35:171]
# photo_hist = cv2.calcHist([cv2.cvtColor(im1, cv2.COLOR_BGR2GRAY)], [0], None, [256], [0,256])
# gray_card_im = cv2.cvtColor(im2, cv2.COLOR_BGR2GRAY)
# image_hist = cv2.calcHist([gray_card_im], [0], None, [256], [0,256])
score = 0
r = range(0,im1.shape[-1])
for i in r:
photo_hist = cv2.calcHist([im1], [i], None, [256], [0,256])
image_hist = cv2.calcHist([im2], [i], None, [256], [0,256])
score += cv2.compareHist(photo_hist, image_hist, method = cv2.HISTCMP_INTERSECT)
if method == "cor":
im1 = cv2.resize(im1, (im2.shape[1], im2.shape[0]))
# im1 = cv2.cvtColor(im1, cv2.COLOR_BGR2GRAY)
# im2 = cv2.cvtColor(im2, cv2.COLOR_BGR2GRAY)
score = cv2.matchTemplate(im2, im1, cv2.TM_CCOEFF_NORMED)
# score = self.correlation_coefficient(im1, im2)
if method == "diff":
im1 = cv2.resize(im1, (im2.shape[1], im2.shape[0]))
diff = im1 - im2
matrix = np.array(diff)
flat = matrix.flatten()
numchange = np.count_nonzero(flat)
score = 100 * float(numchange) / float(len(flat))
if method == "hog":
im1 = cv2.resize(im1, (im2.shape[1], im2.shape[0]))
im1 = cv2.cvtColor(im1, cv2.COLOR_BGR2GRAY)
im2 = cv2.cvtColor(im2, cv2.COLOR_BGR2GRAY)
# H1 = feature.hog(im1, orientations=9, pixels_per_cell=(8, 8), cells_per_block=(2, 2), transform_sqrt=True)
# H2 = feature.hog(im2, orientations=9, pixels_per_cell=(8, 8), cells_per_block=(2, 2), transform_sqrt=True)
# score = cv2.compareHist(np.float32(H1), np.float32(H2), method = cv2.HISTCMP_BHATTACHARYYA)
im1 = np.float32(im1) / 255.0
im2 = np.float32(im2) / 255.0
# Calculate gradient
im1_gx = cv2.Sobel(im1, cv2.CV_32F, 1, 0, ksize=1)
im1_gy = cv2.Sobel(im1, cv2.CV_32F, 0, 1, ksize=1)
im2_gx = cv2.Sobel(im2, cv2.CV_32F, 1, 0, ksize=1)
im2_gy = cv2.Sobel(im2, cv2.CV_32F, 0, 1, ksize=1)
# Python Calculate gradient magnitude and direction ( in degrees )
mag1, angle1 = cv2.cartToPolar(im1_gx, im1_gy, angleInDegrees=True)
mag2, angle2 = cv2.cartToPolar(im2_gx, im2_gy, angleInDegrees=True)
# Compute corelation between angles
# (h, w) = angle1.shape[:2]
# print(h)
# print(w)
score = np.nanmin(1 - scipy.spatial.distance.cdist(angle1, angle2, "cosine"))
# score = self.ccoeff_normed(angle1, angle2)
if method == "orb":
# See https://docs.opencv.org/3.0-beta/doc/py_tutorials/py_feature2d/py_matcher/py_matcher.html
im1 = cv2.cvtColor(im1, cv2.COLOR_BGR2GRAY)
im2 = cv2.imread(im2, 0)
#im2 = cv2.cvtColor(im2, cv2.COLOR_BGR2GRAY)
# Initiate SIFT detector
orb = cv2.ORB_create()
# find the keypoints and descriptors with SIFT
kp1, des1 = orb.detectAndCompute(im1,None)
kp2, des2 = orb.detectAndCompute(im2,None)
logging.info(des1.shape)
logging.info(des2.shape)
logging.info(type(des1))
logging.info(type(des2))
# create BFMatcher object
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
# Match descriptors.
matches = bf.match(des1,des2)
matches = sorted(matches, key = lambda x:x.distance)
score = sum(m.distance for m in matches[:20])
# for m in matches[:20]:
# score += m.distance
return score