def LumConDrift(bgImg,fundusMask):
m,n = bgImg.shape
tsize = 50
indx=0
indy=0
i = tsize
ldrift = np.zeros((int(m/tsize),int(n/tsize)),np.float)
cdrift = np.zeros((int(m/tsize),int(n/tsize)),np.float)
while(i<m):
j = tsize
while(j<n):
if (i+tsize>=m and j+tsize<n):
block = bgImg[i-tsize:m, j-tsize:j+tsize]
elif (i+tsize<m and j+tsize>=n):
block = bgImg[i-tsize:i+tsize, j-tsize:n]
elif (i+tsize>=m and j+tsize>=n):
block = bgImg[i-tsize:m, j-tsize:n]
else :
block = bgImg[i-tsize:i+tsize, j-tsize:j+tsize]
mean,std = cv2.meanStdDev(block)
ldrift[indx,indy] = mean
cdrift[indx,indy] = std
indy = indy+1
j = j+tsize
indy = 0
indx = indx+1
i = i+tsize
ldrift = cv2.resize(ldrift,(n,m),interpolation = cv2.INTER_CUBIC)
cdrift = cv2.resize(cdrift,(n,m),interpolation = cv2.INTER_CUBIC)
ldrift = cv2.multiply(ldrift,fundusMask.astype(float))
cdrift = cv2.multiply(cdrift,fundusMask.astype(float))
return ldrift,cdrift
def emphasis(self, img, scale=2, fraction=0.5) :
"""Return monochrome image with doubled(scale) selected color
minus half of the other two colors(/fraction) added together.
Where color is Blue (0), Green (1-default), or Red (2)"""
return cv2.subtract(cv2.multiply(img[:,:,self.color],scale),
cv2.multiply(cv2.add( img[:,:,(self.color+1)%3],
img[:,:,(self.color+2)%3]),fraction))
def online_variance(new_data,curr_var,curr_iter,curr_mean): if curr_iter==1: new_mean = new_data; new_var = cv2.multiply(new_data,0); return new_mean,new_var; else: pa=cv2.subtract(new_data,curr_mean); pa=cv2.divide(pa,curr_iter,1); new_mean=cv2.add(pa,curr_mean); #new_mean = curr_mean + (new_data - curr_mean)/curr_iter; prev_S = cv2.multiply(curr_var,(curr_iter - 2)); # pd1=cv2.subtract(new_data,curr_mean); pd2=cv2.subtract(new_data,new_mean); pd=cv2.multiply(pd1,pd2); new_S=cv2.add(pd,prev_S); #new_S = prev_S + (new_data - curr_mean) .* (new_data - new_mean); new_var=cv2.divide(new_S,curr_iter-1); #new_var = new_S/(curr_iter - 1); return (new_mean),(new_var);
def do(self, img):
dstImage = img
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
(minVal, maxVal, minLoc, maxLoc) = cv2.minMaxLoc(gray)
cv2.multiply(cv2.add(img, -minVal), 255 / (maxVal - minVal), dstImage)
return dstImage
def emphasis(self, img, color) :
"""Create a monochrome image by doubling the selected color
and subtracting half of the other two colors added together.
Where color is Blue (0), Green (1), or Red (2)"""
return cv2.subtract(cv2.multiply(img[:,:,color],2),
cv2.multiply(cv2.add( img[:,:,(color+1)%3]/3,
img[:,:,(color+2)%3]/3),0.5))
def apply_new_face(self, image, new_face, image_mask, mat, image_size, size):
base_image = numpy.copy( image )
new_image = numpy.copy( image )
cv2.warpAffine( new_face, mat, image_size, new_image, cv2.WARP_INVERSE_MAP | cv2.INTER_CUBIC, cv2.BORDER_TRANSPARENT )
outImage = None
if self.seamless_clone:
unitMask = numpy.clip( image_mask * 365, 0, 255 ).astype(numpy.uint8)
maxregion = numpy.argwhere(unitMask==255)
if maxregion.size > 0:
miny,minx = maxregion.min(axis=0)[:2]
maxy,maxx = maxregion.max(axis=0)[:2]
lenx = maxx - minx;
leny = maxy - miny;
masky = int(minx+(lenx//2))
maskx = int(miny+(leny//2))
outimage = cv2.seamlessClone(new_image.astype(numpy.uint8),base_image.astype(numpy.uint8),unitMask,(masky,maskx) , cv2.NORMAL_CLONE )
return outimage
foreground = cv2.multiply(image_mask, new_image.astype(float))
background = cv2.multiply(1.0 - image_mask, base_image.astype(float))
outimage = cv2.add(foreground, background)
return outimage
def find_marker(image, red_thres, green_thres, sat_thres):
# h,w, channels = img.shape
# get red and sat
hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
blue, green, red = cv2.split(image)
hue, sat, val = cv2.split(hsv)
# find the marker by looking for red, with high saturation
sat = cv2.inRange(sat, np.array((sat_thres[0])), np.array((sat_thres[1])))
red = cv2.inRange(red, np.array((red_thres[0])), np.array((red_thres[1])))
green = cv2.inRange(green, np.array((green_thres[0])), np.array((green_thres[1])))
# AND the two thresholds, finding the car
car = cv2.multiply(red, sat)
car = cv2.multiply(car, green)
# remove noise (not doing it now because the POIs are very small)
# elem = cv2.getStructuringElement(cv2.MORPH_RECT,(3,3))
# car = cv2.erode(car,elem, iterations=1)
# car = cv2.dilate(car,elem, iterations=3)
# return cv2.boundingRect(car)
img, contours, hierarchy = cv2.findContours(car.copy(), cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# import ipdb; ipdb.set_trace()
# return cv2.boundingRect(contours[1])
return map(lambda x: cv2.boundingRect(x), contours)
def ColormapBoundry(image, mean, std, low, high):
newMin = mean - 0.5*(8-low)*std
newMax = mean + 0.5*high*std
newSlope = 255.0/(newMax-newMin)
cv2.subtract(image, newMin, image)
cv2.multiply(image, newSlope, image)
return image.astype("uint8", copy=False)
def draw_mouth(mouth, character, x, y, width, height):
fit_image = fit_character(mouth[0], width, height)
fit_mask = fit_character(mouth[1], width, height)
fit_mask2 = fit_character(mouth[2], width, height)
fit_height, fit_width = fit_image.shape[0:2]
y_offset = y + fit_height / 6
y_offset = max(0, min(y_offset, character.shape[0] - fit_height))
x_offset = x + (width - fit_width) / 2
x_offset = max(0, min(x_offset, character.shape[1] - fit_width))
y0, y1 = y_offset, (y_offset+fit_height)
x0, x1 = x_offset, (x_offset+fit_width)
fit_mask = numpy.float32(fit_mask) / 255.0
fit_mask2 = numpy.float32(fit_mask2) / 255.0
char_region = numpy.float32(character[y0:y1,x0:x1])
inverse_fit_mask = fit_mask * -1 + 1.0
mul = cv2.multiply(char_region, fit_mask)
m1 = cv2.mean(mul)
m2 = cv2.mean(fit_mask)
avg = numpy.float32(map(lambda x, y: x/(y * 255.0) if y else 0.0, m1, m2))
r = numpy.ones((fit_width,fit_height),numpy.float32) * avg[0]
g = numpy.ones((fit_width,fit_height),numpy.float32) * avg[1]
b = numpy.ones((fit_width,fit_height),numpy.float32) * avg[2]
rgb = cv2.merge((r,g,b))
rgb += (rgb * -1.0 + 0.8) * fit_mask2
fit_image = cv2.multiply(numpy.float32(fit_image), rgb)
fit_image = cv2.multiply(fit_image, inverse_fit_mask)
character[y0:y1,x0:x1] = numpy.uint8(mul + fit_image)
def contrast(image) : (ret,img) = cv2.threshold( cv2.add(cv2.multiply( cv2.add(cv2.multiply( cv2.add(cv2.multiply(image,2.0),-60) ,2.0),-60) ,2.1),-100), 127,255,cv2.THRESH_BINARY) return img
def overlay_on(self, frame):
float_frame = frame.astype(float)
float_frame[self.get_valid_backdrop_slices_for_shape(frame.shape)] = \
cv2.multiply(1.0 - self.alpha_channel[self.get_valid_sprite_slices_for_shape(frame.shape)],
float_frame[self.get_valid_backdrop_slices_for_shape(frame.shape)])
float_frame[self.get_valid_backdrop_slices_for_shape(frame.shape)] += \
cv2.multiply(self.alpha_channel, self.image.astype(float))[self.get_valid_sprite_slices_for_shape(frame.shape)]
return float_frame.astype(np.uint8)
def build_filters(sigma, gamma, ksize, thetaRange, lmdaRange, thetaDivider, lmdaDivider):
filters = []
for theta in np.arange(0, thetaRange, thetaRange / thetaDivider):
for lmda in np.arange(1, lmdaRange, lmdaDivider):
kern = cv2.getGaborKernel((ksize, ksize), sigma, theta, lmda, gamma, ktype=cv2.CV_64F)
cv2.multiply(kern, 1 * kern, kern)
filters.append(kern)
return filters
def compute_DELTAE(imagename1,imagename2):
img1 = cv2.imread(imagename1)
img2 = cv2.imread(imagename2)
img1 = cv2.cvtColor(img1, cv2.COLOR_RGB2LAB)
img2 = cv2.cvtColor(img2, cv2.COLOR_RGB2LAB)
#cv2.imwrite(imagename2+".png",img1)
# s1 = cv2.absdiff(img1,img2)
# s1 = np.float32(s1)
# s1 = cv2.multiply(s1,s1)
# s1 = cv2.sqrt(s1)
L1,a1,b1 = cv2.split(img1)
L2,a2,b2 = cv2.split(img2)
dL = L1 - L2
da = a1-a2
db = b1-b2
# cv2.imwrite(imagename2+".png",dL)
# dL_2 = cv2.multiply(dL,dL)
# da_2 = cv2.multiply(da,da)
# db_2 = cv2.multiply(db,db)
# dL_2 = np.float32(dL_2)
# da_2 = np.float32(da_2)
# db_2 = np.float32(db_2)
# dE = cv2.sqrt( (dL_2) + (da_2) + (db_2))
# mde = cv2.mean(dE)
# print mde
c1 = np.sqrt(cv2.multiply(a1,a1) + cv2.multiply(b1,b1))
c2 = np.sqrt(cv2.multiply(a2,a2) + cv2.multiply(b2,b2))
dCab = c1-c2
dH = np.sqrt(cv2.multiply(da,da) + cv2.multiply(db,db)- cv2.multiply(db,db))
sL = 1
K1 = 0.045 #can be changed
K2 = 0.015 #can be changed
sC = 1+K1*c1
sH = 1+K2 *c1
kL = 1 #can be changed
t1 = cv2.divide(dL,kL*sL)
t2 = cv2.divide(dCab,sC)
t3 = cv2.divide(dH,sH)
t1 = cv2.multiply(t1,t1)
t2 = cv2.multiply(t2,t2)
t3 = cv2.multiply(t3,t3)
t1 = np.float32(t1)
t2 = np.float32(t2)
t3 = np.float32(t3)
dE = cv2.sqrt(t1+t2+t3)
mde = cv2.mean(dE)
return "{0:.4f}".format(mde[0])
def get_object_grass(image):
# outputx = cv2.resize(output,(649,486))
# cv2.imshow('image',outputx)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
B = image[:,:,0]
G = image[:,:,1]
R = image[:,:,2]
R_ = cv2.multiply(R, 0.25)
G = cv2.multiply(G, 0.25)
B = cv2.multiply(B, 0.5)
R = cv2.subtract(R, R_)
R = cv2.subtract(R, G)
new_image = cv2.subtract(R, B)
#gray = cv2.add(cv2.add(image[:,:,2],-image[:,:,1]/2),-image[:,:,0]/2)
# output = cv2.resize(new_image,(649,486))
# cv2.imshow('image',output)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
gray = cv2.medianBlur(new_image,3)
#gray = cv2.bilateralFilter(gray,5,75,75)
# find the contours in the edged image and keep the largest one;
_, cnts, _ = cv2.findContours(gray.copy(), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
cnts = sorted(cnts, key = cv2.contourArea, reverse = True)[:10]
for c in cnts:
# approximate the contour
marker = cv2.minAreaRect(c)
x,y,w,h = cv2.boundingRect(c)
coords = (x,y,w,h)
M = cv2.moments(c)
cx,cy = int(M['m10']/M['m00']), int(M['m01']/M['m00'])
center = (cx,cy)
colors = get_colors(image,coords)
#print(marker)
area = marker[1][0]*marker[1][1]
if ((abs(marker[2]) > 60 or abs(marker[2]) < 40) and area > 7000 and area < 250000
and marker[1][0]> 50 and marker[1][1] > 50 and colors['blue']<200
):
return (marker,coords,center)
else:
continue
# if our approximated contour has four points, then
# we can assume that we have found our screen
return (((0, 0), (0, 0), 0),(0,0,0,0),(0,0))
def doMDScalibration(debug):
retVal = (0, 0)
cap = cv2.VideoCapture(1)
cap.set(10,-0.5) #set brightness
cap.set(12,0.8) #set brightness
ret, frame = cap.read()
#unsharp mask
unsharp_mask = cv2.blur(frame, (2, 2))
frame = cv2.addWeighted(frame, 1.5, unsharp_mask, -0.5, 0.0)
#contrast enchancement
array_alpha = np.array([1.2])
array_beta = np.array([-30.0])
# add a beta value to every pixel
cv2.add(frame, array_beta, frame)
# multiply every pixel value by alpha
cv2.multiply(frame, array_alpha, frame)
boundaries = [([0, 150, 180], [10, 205, 230])] #very rough color estimation, no absolute color detection
# loop over the boundaries which actually doesn't matter right now, it only runs once
for (lower, upper) in boundaries:
# create NumPy arrays from the boundaries
lower = np.array(lower, dtype = "uint8")
upper = np.array(upper, dtype = "uint8")
# find the colors within the specified boundaries and apply
# the mask
mask = cv2.inRange(frame, lower, upper)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(6,8))
mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(5,5))
mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel)
output = cv2.bitwise_and(frame, frame, mask = mask)
Omoments = cv2.moments(mask)
dM01 = Omoments['m01']
dM10 = Omoments['m10']
dArea = Omoments['m00']
if dArea > 8000: #the dot is on the screen
posX = int(dM10/dArea)
posY = int(dM01/dArea)
retVal = (posX, posY)
#print a circle if the indicator was detected
cv2.circle(frame, (posX, posY), 8, (0, 255, 255), -1)
if(debug):
#debug for showing the image to ensure the puck is detected
cv2.imshow("images", np.hstack([frame, output]))
cv2.waitKey(0)
cap.release()
cv2.destroyAllWindows()
print retVal
return retVal
def test_umat_construct(self):
data = np.random.random([512, 512])
# UMat constructors
data_um = cv2.UMat(data) # from ndarray
data_sub_um = cv2.UMat(data_um, [0, 256], [0, 256]) # from UMat
data_dst_um = cv2.UMat(256, 256, cv2.CV_64F) # from size/type
# simple test
cv2.multiply(data_sub_um, 2., dst=data_dst_um)
assert np.allclose(2. * data[:256, :256], data_dst_um.get())
def get_blurred_field(video, upper_left, lower_right, blur, brightness):
middle_mat = np.zeros(video.shape, dtype=np.float32)
cv2.rectangle(middle_mat, upper_left, lower_right, (1.0, 1, 1), thickness=-1)
middle_mat = cv2.blur(middle_mat, (20, 20))
middle_field_vid = cv2.blur(video, (int(blur), int(blur)))
middle_field_vid = cv2.multiply(middle_field_vid, brightness * middle_mat, dtype=3)
rest_vid = cv2.multiply(video, 1 - middle_mat, dtype=3)
return cv2.add(middle_field_vid, rest_vid)
def scaleMul(img, image,wavelet):
print '1'
cA1,cH1,cV1,cD1= WaveletTransform(image,wavelet)
print '2'
cA2,cH2,cV2,cD2= WaveletTransform(cA1,wavelet)
print '3'
cA3,cH3,cV3,cD3= WaveletTransform(cA2,wavelet)
print '4'
cA4,cH4,cV4,cD4= WaveletTransform(cA3,wavelet)
cHH1=cv2.multiply(cH1,cH2)
cVV1=cv2.multiply(cV1,cV2)
cDD1=cv2.multiply(cD1,cD2)
cHH2=cv2.multiply(cH2,cH3)
cVV2=cv2.multiply(cV2,cV3)
cDD2=cv2.multiply(cD2,cD3)
cHH3=cv2.multiply(cH3,cH4)
cVV3=cv2.multiply(cV3,cV4)
cDD3=cv2.multiply(cD3,cD4)
#####------Adding the horizontal, vertical and diagonal details to form a combined edge map------#####
final1=cHH1+cVV1+cDD1
final2=cHH2+cVV2+cDD2
final3=cHH3+cVV3+cDD3
imsave('results/'+img+'level_1and2.png',final1)
imsave('results/'+img+'level_2and3.png',final2)
imsave('results/'+img+'level_3and4.png',final3)
def imageenhancement(image):
hsv=cv2.cvtColor(image,cv2.COLOR_BGR2HSV)
bgImg,fundusMask = computebgimg(image)
bgImg = cv2.multiply(image[:,:,1].astype(float),bgImg.astype(float))
ldrift,cdrift = LumConDrift(bgImg,fundusMask)
g = image[:,:,1].astype(float)
imgCorr = cv2.divide(cv2.subtract(g,ldrift),(cv2.add(cdrift,0.0001)))
imgCorr = cv2.multiply(imgCorr,fundusMask.astype(float))
imgCorr = cv2.add(imgCorr,np.abs(np.min(imgCorr)))
imgCorr = cv2.divide(imgCorr,np.max(imgCorr))
imgCorr = cv2.multiply(imgCorr,fundusMask.astype(float))
image = image.astype(float)
image[:,:,0] = cv2.divide(cv2.multiply(imgCorr,image[:,:,0]),hsv[:,:,2].astype(float))
image[:,:,1] = cv2.divide(cv2.multiply(imgCorr,image[:,:,1]),hsv[:,:,2].astype(float))
image[:,:,2] = cv2.divide(cv2.multiply(imgCorr,image[:,:,2]),hsv[:,:,2].astype(float))
fundusMask = fundusMask.astype(float)
image[:,:,0] = cv2.multiply(image[:,:,0],fundusMask)
image[:,:,1] = cv2.multiply(image[:,:,1],fundusMask)
image[:,:,2] = cv2.multiply(image[:,:,2],fundusMask)
out = image[:,:,1]*255
return out
def process_frame(frame):
""" Process frame based on user input """
center_val = cv2.getTrackbarPos(tbar_hue_select_name, win_debug_name)
span_val = cv2.getTrackbarPos(tbar_span_name, win_debug_name)
#get mirror point from values
offset_val = 255/2 - center_val
#get bounds
#convert to hsv
frame_hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
#convert grayscale
frame = get_channel(frame_hsv, 0)
#setupe debug window
temp = cv2.resize(frame,None,fx=3,fy=1,interpolation=cv2.INTER_NEAREST)
(h,w) = temp.shape
x1 = 0
x2 = w/3
x3 = 2* w/3
output_frame = np.zeros(temp.shape+(3,), np.uint8)
#print input frame (#1)
output_frame[0:h,0:x2] = cv2.cvtColor(cv2.multiply(2,frame), cv2.COLOR_GRAY2BGR) ####################
#print rough threshold frame (#2)
output_frame[0:h,x2:x3] = cv2.cvtColor(
cv2.inRange(frame,center_val-span_val/2,center_val+span_val/2),
cv2.COLOR_GRAY2BGR
) #############################
#normalize
normalized_hue_frame = frame
normalized_hue_frame += offset_val
if ABS_DIFF:
distance_frame = cv2.absdiff(normalized_hue_frame,127)
normalized_hue_frame = cv2.subtract(127,distance_frame)
#print final frame (#3)
output_frame[0:h,x3:w] = cv2.cvtColor(
cv2.multiply(2,normalized_hue_frame),
cv2.COLOR_GRAY2BGR) ################
#frame_hsv[:,:,0] = normalized_hue_frame
#output_frame[0:h,x3:w] = cv2.cvtColor(frame_hsv,cv2.COLOR_HSV2BGR) ################
print("Hue Shift: frame += %d. Hue Sel = %d +/- %d" % (offset_val,center_val,span_val/2))
return output_frame
def test_umat_construct(self):
data = np.random.random([512, 512])
# UMat constructors
data_um = cv.UMat(data) # from ndarray
data_sub_um = cv.UMat(data_um, (128, 256), (128, 256)) # from UMat
data_dst_um = cv.UMat(128, 128, cv.CV_64F) # from size/type
# test continuous and submatrix flags
assert data_um.isContinuous() and not data_um.isSubmatrix()
assert not data_sub_um.isContinuous() and data_sub_um.isSubmatrix()
# test operation on submatrix
cv.multiply(data_sub_um, 2., dst=data_dst_um)
assert np.allclose(2. * data[128:256, 128:256], data_dst_um.get())
def change_constrast(image, alpha, beta):
"""
Change the contrast of image, return processed image
Input: image, contrast coefficient and brightness coefficient
"""
array_alpha = np.array([alpha])
array_beta = np.array([beta])
cv2.add(image, array_beta, image)
cv2.multiply(image, array_alpha, image)
return image
def detect_blurry_faces(self, face, t_mat, resized_image, filename):
""" Detect and move blurry face """
if not hasattr(self.args, 'blur_thresh') or not self.args.blur_thresh:
return None
blurry_file = None
aligned_landmarks = self.extractor.transform_points(
face.landmarks_as_xy(),
t_mat,
256,
48)
feature_mask = self.extractor.get_feature_mask(aligned_landmarks / 256,
256,
48)
feature_mask = cv2.blur(feature_mask, (10, 10))
isolated_face = cv2.multiply(
feature_mask,
resized_image.astype(float)).astype(np.uint8)
blurry, focus_measure = is_blurry(isolated_face, self.args.blur_thresh)
if blurry:
print("{}'s focus measure of {} was below the blur threshold, "
"moving to \"blurry\"".format(Path(filename).stem,
focus_measure))
blurry_file = get_folder(Path(self.output_dir) /
Path("blurry")) / Path(filename).stem
return blurry_file
def __tutorial_hough_circle_detection_cv(img_path, min_dim=40, max_dim=60):
img_color = cv2.imread(img_path)
img_filtered = cv2.pyrMeanShiftFiltering(img_color, 10, 10)
img_filtered = cv2.cvtColor(img_filtered, cv2.COLOR_BGRA2GRAY)
img_filtered = img_filtered.astype(float)
img_blurred = cv2.GaussianBlur(img_filtered, (7, 7), sigmaX=0)
img_laplacian = cv2.Laplacian(img_blurred, ddepth=cv2.CV_64F)
weight = 0.01 * 40
scale = 0.01 * 20
img_sharpened = (1.5 * img_filtered) - (0.5 * img_blurred) - (weight * cv2.multiply(img_filtered, scale * img_laplacian))
img_sharpened = img_sharpened.astype("uint8")
min_r = int(min_dim / 2)
max_r = int(max_dim / 2)
circles = cv2.HoughCircles(img_sharpened, cv2.HOUGH_GRADIENT, 1, 1, param1=50, param2=30, minRadius=min_r, maxRadius=max_r)
if circles is not None:
circles = np.around(circles).astype(int)
for i in circles[0, :]:
# draw the outer circle
cv2.circle(img_color, (i[0], i[1]), i[2], (0, 255, 0), 2)
# draw the center of the circle
cv2.circle(img_color, (i[0], i[1]), 2, (0, 0, 255), 3)
cv2.imwrite("D://_Dataset//GTSDB//Test_Regions//_img2_1.png", img_sharpened)
cv2.imwrite("D://_Dataset//GTSDB//Test_Regions//_img2_2.png", img_color)
def tantriggs(image):
# Convert to float
image = np.float32(image)
image = cv2.pow(image, GAMMA)
image = difference_of_gaussian(image)
# mean 1
tmp = cv2.pow(cv2.absdiff(image, 0), ALPHA)
mean = cv2.mean(tmp)[0]
image = cv2.divide(image, cv2.pow(mean, 1.0/ALPHA))
# mean 2
tmp = cv2.pow(cv2.min(cv2.absdiff(image, 0), TAU), ALPHA)
mean = cv2.mean(tmp)[0]
image = cv2.divide(image, cv2.pow(mean, 1.0/ALPHA))
# tanh
exp_x = cv2.exp(cv2.divide(image, TAU))
exp_negx = cv2.exp(cv2.divide(-image, TAU))
image = cv2.divide(cv2.subtract(exp_x, exp_negx), cv2.add(exp_x, exp_negx))
image = cv2.multiply(image, TAU)
image = cv2.normalize(image, None, 0, 255, cv2.NORM_MINMAX, cv2.CV_8UC1)
return image
def ContrastCorrection(img, correction):
x,y=img.shape
bright= np.ndarray( shape=(x,y), dtype="uint8" )
bright.fill(2)
img = cv2.multiply(img, bright)
return img
def process(self, output_item):
""" Detect and move blurry face """
extractor = AlignerExtract()
for idx, detected_face in enumerate(output_item["detected_faces"]):
frame_name = detected_face["file_location"].parts[-1]
face = detected_face["face"]
logger.trace("Checking for blurriness. Frame: '%s', Face: %s", frame_name, idx)
aligned_landmarks = face.aligned_landmarks
resized_face = face.aligned_face
size = face.aligned["size"]
padding = int(size * 0.1875)
feature_mask = extractor.get_feature_mask(
aligned_landmarks / size,
size, padding)
feature_mask = cv2.blur( # pylint: disable=no-member
feature_mask, (10, 10))
isolated_face = cv2.multiply( # pylint: disable=no-member
feature_mask,
resized_face.astype(float)).astype(np.uint8)
blurry, focus_measure = self.is_blurry(isolated_face)
if blurry:
blur_folder = detected_face["file_location"].parts[:-1]
blur_folder = get_folder(Path(*blur_folder) / Path("blurry"))
detected_face["file_location"] = blur_folder / Path(frame_name)
logger.verbose("%s's focus measure of %s was below the blur threshold, "
"moving to 'blurry'", frame_name, "{0:.2f}".format(focus_measure))
def dropout(i,p):
mask = np.empty(i.shape, dtype=np.int16)
mask = cv2.randu(mask,0,255)
val, mask = cv2.threshold(mask,p*255,255,cv2.THRESH_BINARY)
mask = np.asarray(mask,dtype=np.float64) / 255.0
return cv2.multiply(i,mask)
def blurshift(img_blur,sigma0,rhoi,phii):
img_blur = cv2.cvtColor(img_blur, cv2.COLOR_BGR2GRAY)
rows,cols = img_blur.shape
alpha = 1
#the std deviation of the guassian kernal
sigma1 = int(sigma0)+alpha*(int(rhoi))
##img_blur = cv2.GaussianBlur(img,(sigma1,sigma1),0)
filter_gaussian = gauss2D((rows,cols),sigma1)
##plt.imshow(filter_gaussian, cmap=plt.get_cmap('jet'), interpolation='nearest')
##plt.colorbar()
##plt.show()
# the shift in x and y direction for DOG response
delta_xi = -1*int(rhoi)*np.cos(int(phii))
delta_yi = -1*int(rhoi)*np.sin(int(phii))
x1 = sigma1
y1 = sigma1
rows1,cols1 =img_blur.shape
M = np.float32([[1,0,delta_xi+x1],[0,1,delta_yi+y1]])
#translating the response
Dogresp_trans = cv2.warpAffine(img_blur,M,(cols1,rows1))
#mulitplying the coefficients of gaussian and translated DOG response
Dog_blur_shifted = cv2.multiply(Dogresp_trans.astype(float),filter_gaussian)
return Dog_blur_shifted
def thresholdColor(img, colattr):
(domchan, dommin, first, second) = colattr
channels = cv2.split(img)#red, green, blue
width, height, cha = img.shape
mult = np.empty((width,height)).astype(np.uint8)
mult.fill(255)
red = channels[2].astype(np.uint8)
green = channels[1].astype(np.uint8)
blue = channels[0].astype(np.uint8)
firsttype = np.zeros(img.shape)
secondtype = np.zeros(img.shape)
if domchan == "r":
zerotype = (red > dommin)
firsttype = np.true_divide(red,green)#r/g
secondtype = np.true_divide(red,blue)#r/b
elif domchan == "g":
zerotype = (green > dommin)
firsttype = np.true_divide(green,red)#g/r
secondtype = np.true_divide(green,blue)#g/b
zerotype = zerotype.astype(np.int)
firsttype = (firsttype > first).astype(np.int)# & (firsttype < first[1])
secondtype = (secondtype > second).astype(np.int)# & (secondtype < second[1])
combined = cv2.bitwise_and(cv2.bitwise_and(zerotype, secondtype), firsttype)
combined = cv2.multiply(combined.astype(np.uint8), mult)
return combined
def process_output(image, output, raw_img, threshold, min_area):
height, width, _ = image.shape
pred_x = output[0, :]
pred_y = output[1, :]
magnitude, angle = cv2.cartToPolar(pred_x, pred_y)
thr = np.array([threshold])
mask = cv2.compare(magnitude, thr, cv2.CMP_GT)
mask = mask.astype(np.float32)
parent = np.zeros((height, width, 2), np.float32)
ending = np.zeros((height, width), np.float32)
merged_ending = np.zeros((height, width), np.float32)
PI = np.pi
for row in range(0, height):
for col in range(0, width):
if mask[row][col] == 255:
if angle[row][col] < PI/8 or angle[row][col] >= 15*PI/8:
parent[row][col] = [1,0]
if row+1 <= height-1 and mask[row+1][col] == 0:
ending[row][col] = 1
elif angle[row][col] >= PI/8 and angle[row][col] < 3*PI/8:
parent[row][col] = [1,1]
if row+1 <= height-1 and col+1 <= width-1 and mask[row+1][col+1] == 0:
ending[row][col] = 1
elif angle[row][col] >= 3*PI/8 and angle[row][col] < 5*PI/8:
parent[row][col] = [0,1]
if col+1 <= width-1 and mask[row][col+1] == 0:
ending[row][col] = 1
elif angle[row][col] >= 5*PI/8 and angle[row][col] < 7*PI/8:
parent[row][col] = [-1,1]
if row-1 >= 0 and col+1 <= width-1 and mask[row-1][col+1] == 0:
ending[row][col] = 1
elif angle[row][col] >= 7*PI/8 and angle[row][col] < 9*PI/8:
parent[row][col] = [-1,0]
if row-1 >= 0 and mask[row-1][col] == 0:
ending[row][col] = 1
elif angle[row][col] >= 9*PI/8 and angle[row][col] < 11*PI/8:
parent[row][col] = [-1,-1]
if row-1 >= 0 and col-1 >= 0 and mask[row-1][col-1] == 0:
ending[row][col] = 1
elif angle[row][col] >= 11*PI/8 and angle[row][col] < 13*PI/8:
parent[row][col] = [0,-1]
if col-1 >= 0 and mask[row][col-1] == 0:
ending[row][col] = 1
elif angle[row][col] >= 13*PI/8 and angle[row][col] < 15*PI/8:
parent[row][col] = [1,-1]
if row+1 <= height-1 and col-1 >= 0 and mask[row+1][col-1] == 0:
ending[row][col] = 1
p = Coordinate()
pc = Coordinate()
pt = Coordinate()
visited = np.zeros((height, width), np.float32)
dict_id = np.zeros((height, width, 2), np.float32)
# blob lableing to construct trees encoded by P
# get depth each pixel in text instance
sup_idx = 1
for row in range(0, height):
for col in range(0, width):
if mask[row][col] == 255 and visited[row][col] == 0:
p.x = row
p.y = col
Q = queue.Queue()
Q.put(p)
while not Q.empty():
pc = Q.get()
dict_id[pc.x][pc.y][0] = sup_idx
visited[pc.x][pc.y] = 1
for dx in range(-1, 2):
for dy in range(-1, 2):
pt.x = pc.x + dx
pt.y = pc.y + dy
if pt.x >= 0 and pt.x <= height-1 and pt.y >= 0 and pt.y <= width-1:
if visited[pt.x][pt.y] == 0 and (parent[pt.x][pt.y][0] != 0 or parent[pt.x][pt.y][1] != 0):
if parent[pt.x][pt.y][0] == -1*dx and parent[pt.x][pt.y][1] == -1*dy:
Q.put(pt.copy())
dict_id[pc.x][pc.y][1] = max(
dict_id[pc.x][pc.y][1], dict_id[pt.x][pt.y][1]+1)
elif parent[pc.x][pc.y][0] == 1*dx and parent[pc.x][pc.y][1] == 1*dy:
Q.put(pt.copy())
dict_id[pt.x][pt.y][1] = max(
dict_id[pt.x][pt.y][1], dict_id[pc.x][pc.y][1]+1)
sup_idx += 1
element = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
# fill hole in ending
merged_ending = cv2.dilate(
ending, element, iterations=5).astype(np.float32)
# dilate ending
for row in range(0, height):
for col in range(0, width):
if ending[row][col] == 1:
for dilDepth in range(1, min(int(1*dict_id[row][col][1]-16), 12)+1):
p.x = row+int(parent[row][col][0]*dilDepth)
p.y = col+int(parent[row][col][1]*dilDepth)
if p.x >= 0 and p.x <= height-1 and pt.y >= 0 and pt.y <= width-1:
merged_ending[p.x][p.y] = 1
# find connected Components
cctmp = merged_ending.astype(np.uint8)
# ccnum: num component, cctmp: mask component
ccnum, cctmp = cv2.connectedComponents(
cctmp, connectivity=8, ltype=cv2.CV_16U)
label = cctmp.astype(np.float32)
sup_map_cc = np.zeros((sup_idx), np.int32)
stat = np.zeros((ccnum, 8), np.int32)
# calculate num stat each label and assign label each sup_idx
for row in range(0, height):
for col in range(0, width):
if ending[row][col] == 1:
dx = int(parent[row][col][0])
dy = int(parent[row][col][1])
cc_idx = int(label[row][col])
sup_map_cc[int(dict_id[row][col][0])] = cc_idx
if dx == 1 and dy == 0: # down
stat[cc_idx][0] += 1
if dx == 1 and dy == 1: # down right
stat[cc_idx][1] += 1
if dx == 0 and dy == 1: # right
stat[cc_idx][2] += 1
if dx == -1 and dy == 1: # up right
stat[cc_idx][3] += 1
if dx == -1 and dy == 0: # up
stat[cc_idx][4] += 1
if dx == -1 and dy == -1: # up left
stat[cc_idx][5] += 1
if dx == 0 and dy == -1: # left
stat[cc_idx][6] += 1
if dx == 1 and dy == -1: # down left
stat[cc_idx][7] += 1
cc_map_filted = np.zeros((ccnum), np.int32)
filted_idx = 1
# Filter unblanced Text
for cc_idx in range(1, ccnum):
dif1 = np.abs(stat[cc_idx][0] - stat[cc_idx][4]) # abs(down - up)
# abs(down_right - up_left)
dif2 = np.abs(stat[cc_idx][1] - stat[cc_idx][5])
dif3 = np.abs(stat[cc_idx][2] - stat[cc_idx][6]) # abs(right - left)
# abs(down_left - up_right)
dif4 = np.abs(stat[cc_idx][3] - stat[cc_idx][7])
sum1 = stat[cc_idx][0]+stat[cc_idx][1]+stat[cc_idx][2]+stat[cc_idx][3]
sum2 = stat[cc_idx][4]+stat[cc_idx][5]+stat[cc_idx][6]+stat[cc_idx][7]
difsum = np.abs(sum1-sum2)
sum_total = sum1 + sum2
ratio1 = float(difsum) / float(sum_total)
ratio2 = float(dif1+dif2+dif3+dif4) / float(sum_total)
# keep candidate have low ratio (high opposite directions)
# <=0.6 mean >40% opposite directions
if ratio1 <= 0.6 and ratio2 <= 0.6:
cc_map_filted[cc_idx] = filted_idx
filted_idx += 1
# filter candidate
for row in range(0, height):
for col in range(0, width):
if label[row][col] == 0:
label[row][col] = cc_map_filted[int(
sup_map_cc[int(dict_id[row][col][0])])]
else:
label[row][col] = cc_map_filted[int(label[row][col])]
res = np.zeros((height, width), np.float32)
element_ = cv2.getStructuringElement(cv2.MORPH_RECT, (11, 11))
# get result mask
for i in range(1, filted_idx):
clstmp = cv2.compare(label, np.array([i]), cv2.CMP_EQ)
clstmp = cv2.dilate(clstmp, element_, iterations=1)
clstmp = cv2.erode(clstmp, element_, iterations=1)
clstmp = cv2.compare(clstmp, np.array([0]), cv2.CMP_GT)
clstmp = clstmp.astype(np.float32)
res = cv2.multiply(res, 1-clstmp/255)
res = cv2.add(res, clstmp/255*i)
return seg2bbox(res, cv2.resize(raw_img, (width, height)), raw_img, min_area)
def process_image(self, blur, threshold, adjustment, erode, iterations):
self.img = self.original.copy()
debug_images = []
alpha = float(2.5)
debug_images.append(('Original', self.original))
# Adjust the exposure
exposure_img = cv2.multiply(self.img, np.array([alpha]))
debug_images.append(('Exposure Adjust', exposure_img))
# Convert to grayscale
img2gray = cv2.cvtColor(exposure_img, cv2.COLOR_BGR2GRAY)
debug_images.append(('Grayscale', img2gray))
# Blur to reduce noise
img_blurred = cv2.GaussianBlur(img2gray, (blur, blur), 0)
debug_images.append(('Blurred', img_blurred))
cropped = img_blurred
# Threshold the image
cropped_threshold = cv2.adaptiveThreshold(
cropped, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY,
threshold, adjustment)
debug_images.append(('Cropped Threshold', cropped_threshold))
# Erode the lcd digits to make them continuous for easier contouring
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (erode, erode))
eroded = cv2.erode(cropped_threshold, kernel, iterations=iterations)
debug_images.append(('Eroded', eroded))
# Reverse the image to so the white text is found when looking for the contours
inverse = inverse_colors(eroded)
debug_images.append(('Inversed', inverse))
# Find the lcd digit contours
_, contours, _ = cv2.findContours(
inverse, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE) # get contours
# Assuming we find some, we'll sort them in order left -> right
if len(contours) > 0:
contours, _ = sort_contours(contours)
potential_decimals = []
potential_digits = []
total_digit_height = 0
total_digit_y = 0
# Aspect ratio for all non 1 character digits
desired_aspect = 0.6
# Aspect ratio for the "1" digit
digit_one_aspect = 0.3
# The allowed buffer in the aspect when determining digits
aspect_buffer = 0.15
# Loop over all the contours collecting potential digits and decimals
for contour in contours:
# get rectangle bounding contour
[x, y, w, h] = cv2.boundingRect(contour)
aspect = float(w) / h
size = w * h
# It's a square, save the contour as a potential digit
if size > 100 and aspect >= 1 - .3 and aspect <= 1 + .3:
potential_decimals.append(contour)
# If it's small and it's not a square, kick it out
if size < 20 * 100 and (aspect < 1 - aspect_buffer
and aspect > 1 + aspect_buffer):
continue
# Ignore any rectangles where the width is greater than the height
if w > h:
if self.debug:
cv2.rectangle(self.img, (x, y), (x + w, y + h),
(0, 0, 255), 2)
continue
# If the contour is of decent size and fits the aspect ratios we want, we'll save it
if ((size > 2000 and aspect >= desired_aspect - aspect_buffer
and aspect <= desired_aspect + aspect_buffer) or
(size > 1000 and aspect >= digit_one_aspect - aspect_buffer
and aspect <= digit_one_aspect + aspect_buffer)):
# Keep track of the height and y position so we can run averages later
total_digit_height += h
total_digit_y += y
potential_digits.append(contour)
else:
if self.debug:
cv2.rectangle(self.img, (x, y), (x + w, y + h),
(0, 0, 255), 2)
avg_digit_height = 0
avg_digit_y = 0
potential_digits_count = len(potential_digits)
left_most_digit = 0
right_most_digit = 0
digit_x_positions = []
# Calculate the average digit height and y position so we can determine what we can throw out
if potential_digits_count > 0:
avg_digit_height = float(
total_digit_height) / potential_digits_count
avg_digit_y = float(total_digit_y) / potential_digits_count
if self.debug:
print "Average Digit Height and Y: " + str(
avg_digit_height) + " and " + str(avg_digit_y)
output = ''
ix = 0
# Loop over all the potential digits and see if they are candidates to run through KNN to get the digit
for pot_digit in potential_digits:
[x, y, w, h] = cv2.boundingRect(pot_digit)
# Does this contour match the averages
if h <= avg_digit_height * 1.2 and h >= avg_digit_height * 0.2 and y <= avg_digit_height * 1.2 and y >= avg_digit_y * 0.2:
# Crop the contour off the eroded image
cropped = eroded[y:y + h, x:x + w]
# Draw a rect around it
cv2.rectangle(self.img, (x, y), (x + w, y + h), (255, 0, 0), 2)
debug_images.append(('digit' + str(ix), cropped))
# Call into the KNN to determine the digit
digit = self.predict_digit(cropped)
if self.debug:
print "Digit: " + digit
output += digit
# Helper code to write out the digit image file for use in KNN training
if self.write_digits:
_, full_file = os.path.split(self.file_name)
file_name = full_file.split('.')
crop_file_path = CROP_DIR + '/' + digit + '_' + file_name[
0] + '_crop_' + str(ix) + '.png'
cv2.imwrite(crop_file_path, cropped)
# Track the x positions of where the digits are
if left_most_digit == 0 or x < left_most_digit:
left_most_digit = x
if right_most_digit == 0 or x > right_most_digit:
right_most_digit = x + w
digit_x_positions.append(x)
ix += 1
else:
if self.debug:
cv2.rectangle(self.img, (x, y), (x + w, y + h),
(66, 146, 244), 2)
decimal_x = 0
# Loop over the potential digits and find a square that's between the left/right digit x positions on the
# lower half of the screen
for pot_decimal in potential_decimals:
[x, y, w, h] = cv2.boundingRect(pot_decimal)
if x < right_most_digit and x > left_most_digit and y > (
self.height / 2):
cv2.rectangle(self.img, (x, y), (x + w, y + h), (255, 0, 0), 2)
decimal_x = x
# Once we know the position of the decimal, we'll insert it into our string
for ix, digit_x in enumerate(digit_x_positions):
if digit_x > decimal_x:
# insert
output = output[:ix] + '.' + output[ix:]
break
# Debugging to show the left/right digit x positions
if self.debug:
cv2.rectangle(
self.img, (left_most_digit, int(avg_digit_y)),
(left_most_digit + right_most_digit - left_most_digit,
int(avg_digit_y) + int(avg_digit_height)), (66, 244, 212), 2)
# Log some information
if self.debug:
print "Potential Digits " + str(len(potential_digits))
print "Potential Decimals " + str(len(potential_decimals))
print "String: " + output
return debug_images, output
def farben(hue1, hue2, satu, vis, farbe, titel):
img2 = zuschneiden()
#checken ob Bild richtig zugeschnitten wurde
if (not geklappt):
return
else:
cv2.imshow("zugescnitten", img2)
# Farbkonvertierung
hsv = cv2.cvtColor(img2, cv2.COLOR_BGR2HSV)
# Video in drei Farbkanaele splitten
h, s, v = cv2.split(hsv)
# masken berechnen
h_mask1 = cv2.inRange(h, hue1 - 25, hue2 + 25)
h_mask2 = cv2.inRange(h, hue2 - 25, hue2 + 25)
h_mask = cv2.add(h_mask1, h_mask2)
s_mask = cv2.inRange(s, satu - 25, satu + 25)
v_mask = cv2.inRange(v, vis - 25, vis + 25)
# Multiplikation der Einzelmasken
mask1 = cv2.multiply(v_mask, s_mask)
mask = cv2.multiply(mask1, h_mask)
#Steine rausfiltern und Position bestimmen
contours, hierarchy = cv2.findContours(mask, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
#Steine in der jeweiligen Farbe finden
for index in range(len(contours)):
area = cv2.contourArea(contours[index])
x, y, w, h = cv2.boundingRect(contours[index])
if (w > (w2 * 0.6)) and (h > (h2 * 0.6) or area > swSteine):
x, y, w, h = cv2.boundingRect(contours[index])
cv2.rectangle(mask, (x, y), (x + w, y + h), (255, 255, 225), 3)
if w > (w2 * 1.3):
anzahlNebeneinander = w / w2
print(anzahlNebeneinander)
x_pos = (x + (w * 0.5)) - (10 + w1)
position = 0
if (fl0 < x_pos < fl1):
position = 1
elif (fl1 < x_pos < fl2):
position = 2
elif (fl2 < x_pos < fl3):
position = 3
elif (fl3 < x_pos < fl4):
position = 4
elif (fl4 < x_pos < fl5):
position = 5
elif (fl5 < x_pos < fl6):
position = 6
elif (fl6 < x_pos < fl7):
position = 7
elif (fl7 < x_pos < fl8):
position = 8
#print(position, farbe, x_pos)
#print(fl1,fl2,fl3,fl4,fl5,fl6,fl7,fl8)
sendNoteOn(farbe, position)
#print("jetzt wird das Bild erzeugt")
cv2.imshow(titel, mask)
def roi_objects(img,
roi_type,
roi_contour,
roi_hierarchy,
object_contour,
obj_hierarchy,
device,
debug=None):
"""Find objects partially inside a region of interest or cut objects to the ROI.
Inputs:
img = img to display kept objects
roi_type = 'cutto' or 'partial' (for partially inside)
roi_contour = contour of roi, output from "View and Adjust ROI" function
roi_hierarchy = contour of roi, output from "View and Adjust ROI" function
object_contour = contours of objects, output from "Identifying Objects" function
obj_hierarchy = hierarchy of objects, output from "Identifying Objects" function
device = device number. Used to count steps in the pipeline
debug = None, print, or plot. Print = save to file, Plot = print to screen.
Returns:
device = device number
kept_cnt = kept contours
hierarchy = contour hierarchy list
mask = mask image
obj_area = total object pixel area
:param img: numpy array
:param roi_type: str
:param roi_contour: list
:param roi_hierarchy: list
:param object_contour: list
:param obj_hierarchy: list
:param device: int
:param debug: str
:return device: int
:return kept_cnt: list
:return hierarchy: list
:return mask: numpy array
:return obj_area: int
"""
device += 1
if len(np.shape(img)) == 3:
ix, iy, iz = np.shape(img)
else:
ix, iy = np.shape(img)
size = ix, iy, 3
background = np.zeros(size, dtype=np.uint8)
ori_img = np.copy(img)
w_back = background + 255
background1 = np.zeros(size, dtype=np.uint8)
background2 = np.zeros(size, dtype=np.uint8)
# Allows user to find all objects that are completely inside or overlapping with ROI
if roi_type == 'partial':
for c, cnt in enumerate(object_contour):
length = (len(cnt) - 1)
stack = np.vstack(cnt)
test = []
keep = False
for i in range(0, length):
pptest = cv2.pointPolygonTest(roi_contour[0],
(stack[i][0], stack[i][1]),
False)
if int(pptest) != -1:
keep = True
if keep == True:
if obj_hierarchy[0][c][3] > -1:
cv2.drawContours(w_back,
object_contour,
c, (255, 255, 255),
-1,
lineType=8,
hierarchy=obj_hierarchy)
else:
cv2.drawContours(w_back,
object_contour,
c, (0, 0, 0),
-1,
lineType=8,
hierarchy=obj_hierarchy)
else:
cv2.drawContours(w_back,
object_contour,
c, (255, 255, 255),
-1,
lineType=8,
hierarchy=obj_hierarchy)
kept = cv2.cvtColor(w_back, cv2.COLOR_RGB2GRAY)
kept_obj = cv2.bitwise_not(kept)
mask = np.copy(kept_obj)
obj_area = cv2.countNonZero(kept_obj)
kept_cnt, hierarchy = cv2.findContours(kept_obj, cv2.RETR_TREE,
cv2.CHAIN_APPROX_NONE)
cv2.drawContours(ori_img,
kept_cnt,
-1, (0, 255, 0),
-1,
lineType=8,
hierarchy=hierarchy)
cv2.drawContours(ori_img,
roi_contour,
-1, (255, 0, 0),
5,
lineType=8,
hierarchy=roi_hierarchy)
# Allows user to cut objects to the ROI (all objects completely outside ROI will not be kept)
elif roi_type == 'cutto':
cv2.drawContours(background1,
object_contour,
-1, (255, 255, 255),
-1,
lineType=8,
hierarchy=obj_hierarchy)
roi_points = np.vstack(roi_contour[0])
cv2.fillPoly(background2, [roi_points], (255, 255, 255))
obj_roi = cv2.multiply(background1, background2)
kept_obj = cv2.cvtColor(obj_roi, cv2.COLOR_RGB2GRAY)
mask = np.copy(kept_obj)
obj_area = cv2.countNonZero(kept_obj)
kept_cnt, hierarchy = cv2.findContours(kept_obj, cv2.RETR_TREE,
cv2.CHAIN_APPROX_NONE)
cv2.drawContours(w_back, kept_cnt, -1, (0, 0, 0), -1)
cv2.drawContours(ori_img,
kept_cnt,
-1, (0, 255, 0),
-1,
lineType=8,
hierarchy=hierarchy)
cv2.drawContours(ori_img,
roi_contour,
-1, (255, 0, 0),
5,
lineType=8,
hierarchy=roi_hierarchy)
else:
fatal_error('ROI Type' + str(roi_type) +
' is not "cutto" or "partial"!')
if debug == 'print':
print_image(w_back, (str(device) + '_roi_objects.png'))
print_image(ori_img, (str(device) + '_obj_on_img.png'))
print_image(mask, (str(device) + '_roi_mask.png'))
elif debug == 'plot':
plot_image(w_back)
plot_image(ori_img)
plot_image(mask, cmap='gray')
# print ('Object Area=', obj_area)
return device, kept_cnt, hierarchy, mask, obj_area
def test_arithmetic(self):
npMat1 = np.random.random((128, 128, 3)) - 0.5
npMat2 = np.random.random((128, 128, 3)) - 0.5
cuMat1 = cv.cuda_GpuMat()
cuMat2 = cv.cuda_GpuMat()
cuMat1.upload(npMat1)
cuMat2.upload(npMat2)
cuMatDst = cv.cuda_GpuMat(cuMat1.size(), cuMat1.type())
self.assertTrue(
np.allclose(
cv.cuda.add(cuMat1, cuMat2).download(), cv.add(npMat1,
npMat2)))
cv.cuda.add(cuMat1, cuMat2, cuMatDst)
self.assertTrue(
np.allclose(cuMatDst.download(), cv.add(npMat1, npMat2)))
self.assertTrue(
np.allclose(
cv.cuda.subtract(cuMat1, cuMat2).download(),
cv.subtract(npMat1, npMat2)))
cv.cuda.subtract(cuMat1, cuMat2, cuMatDst)
self.assertTrue(
np.allclose(cuMatDst.download(), cv.subtract(npMat1, npMat2)))
self.assertTrue(
np.allclose(
cv.cuda.multiply(cuMat1, cuMat2).download(),
cv.multiply(npMat1, npMat2)))
cv.cuda.multiply(cuMat1, cuMat2, cuMatDst)
self.assertTrue(
np.allclose(cuMatDst.download(), cv.multiply(npMat1, npMat2)))
self.assertTrue(
np.allclose(
cv.cuda.divide(cuMat1, cuMat2).download(),
cv.divide(npMat1, npMat2)))
cv.cuda.divide(cuMat1, cuMat2, cuMatDst)
self.assertTrue(
np.allclose(cuMatDst.download(), cv.divide(npMat1, npMat2)))
self.assertTrue(
np.allclose(
cv.cuda.absdiff(cuMat1, cuMat2).download(),
cv.absdiff(npMat1, npMat2)))
cv.cuda.absdiff(cuMat1, cuMat2, cuMatDst)
self.assertTrue(
np.allclose(cuMatDst.download(), cv.absdiff(npMat1, npMat2)))
self.assertTrue(
np.allclose(
cv.cuda.compare(cuMat1, cuMat2, cv.CMP_GE).download(),
cv.compare(npMat1, npMat2, cv.CMP_GE)))
cuMatDst1 = cv.cuda_GpuMat(cuMat1.size(), cv.CV_8UC3)
cv.cuda.compare(cuMat1, cuMat2, cv.CMP_GE, cuMatDst1)
self.assertTrue(
np.allclose(cuMatDst1.download(),
cv.compare(npMat1, npMat2, cv.CMP_GE)))
self.assertTrue(
np.allclose(cv.cuda.abs(cuMat1).download(), np.abs(npMat1)))
cv.cuda.abs(cuMat1, cuMatDst)
self.assertTrue(np.allclose(cuMatDst.download(), np.abs(npMat1)))
self.assertTrue(
np.allclose(
cv.cuda.sqrt(cv.cuda.sqr(cuMat1)).download(),
cv.cuda.abs(cuMat1).download()))
cv.cuda.sqr(cuMat1, cuMatDst)
cv.cuda.sqrt(cuMatDst, cuMatDst)
self.assertTrue(
np.allclose(cuMatDst.download(),
cv.cuda.abs(cuMat1).download()))
self.assertTrue(
np.allclose(cv.cuda.log(cv.cuda.exp(cuMat1)).download(), npMat1))
cv.cuda.exp(cuMat1, cuMatDst)
cv.cuda.log(cuMatDst, cuMatDst)
self.assertTrue(np.allclose(cuMatDst.download(), npMat1))
self.assertTrue(
np.allclose(cv.cuda.pow(cuMat1, 2).download(), cv.pow(npMat1, 2)))
cv.cuda.pow(cuMat1, 2, cuMatDst)
self.assertTrue(np.allclose(cuMatDst.download(), cv.pow(npMat1, 2)))
def remove_background(path):
basewidth = 512
img = Image.open(path)
# wpercent = (basewidth / float(img.size[0]))
# hsize = int((float(img.size[1]) * float(wpercent)))
# img = img.resize((basewidth, hsize), Image.ANTIALIAS)
img = img.resize((basewidth, basewidth), Image.ANTIALIAS)
img.save(path)
src = cv2.imread(path, 1)
blurred = cv2.GaussianBlur(src, (5, 5), 0)
blurred_float = blurred.astype(np.float32) / 255.0
edgeDetector = cv2.ximgproc.createStructuredEdgeDetection("model.yml")
edges = edgeDetector.detectEdges(blurred_float) * 255.0
# cv2.imwrite('images/edge-raw.png', edges)
edges_8u = np.asarray(edges, np.uint8)
filterOutSaltPepperNoise(edges_8u)
# cv2.imwrite('images/edge.png', edges_8u)
contour = findSignificantContour(edges_8u)
contourImg = np.copy(src)
cv2.drawContours(contourImg, [contour], 0, (0, 255, 0), 2, cv2.LINE_AA, maxLevel=1)
# cv2.imwrite('images/contour.png', contourImg)
mask = np.zeros_like(edges_8u)
cv2.fillPoly(mask, [contour], 255)
# calculate sure foreground area by dilating the mask
mapFg = cv2.erode(mask, np.ones((5, 5), np.uint8), iterations=10)
# mark inital mask as "probably background"
# and mapFg as sure foreground
trimap = np.copy(mask)
trimap[mask == 0] = cv2.GC_BGD
trimap[mask == 255] = cv2.GC_PR_BGD
trimap[mapFg == 255] = cv2.GC_FGD
# visualize trimap
trimap_print = np.copy(trimap)
trimap_print[trimap_print == cv2.GC_PR_BGD] = 128
trimap_print[trimap_print == cv2.GC_FGD] = 255
# cv2.imwrite('images/trimap.png', trimap_print)
# run grabcut
bgdModel = np.zeros((1, 65), np.float64)
fgdModel = np.zeros((1, 65), np.float64)
rect = (0, 0, mask.shape[0] - 1, mask.shape[1] - 1)
cv2.grabCut(src, trimap, rect, bgdModel, fgdModel, 5, cv2.GC_INIT_WITH_MASK)
# create mask again
mask2 = np.where(
(trimap == cv2.GC_FGD) | (trimap == cv2.GC_PR_FGD),
255,
0
).astype('uint8')
# cv2.imwrite('images/mask2.png', mask2)
contour2 = findSignificantContour(mask2)
mask3 = np.zeros_like(mask2)
cv2.fillPoly(mask3, [contour2], 255)
# blended alpha cut-out
mask3 = np.repeat(mask3[:, :, np.newaxis], 3, axis=2)
mask4 = cv2.GaussianBlur(mask3, (3, 3), 0)
alpha = mask4.astype(float) * 1.1 # making blend stronger
alpha[mask3 > 0] = 255
alpha[alpha > 255] = 255
alpha = alpha.astype(float)
foreground = np.copy(src).astype(float)
foreground[mask4 == 0] = 0
background = np.ones_like(foreground, dtype=float) * 255
# cv2.imwrite('images/foreground.png', foreground)
# cv2.imwrite('images/background.png', background)
# cv2.imwrite('images/alpha.png', alpha)
# Normalize the alpha mask to keep intensity between 0 and 1
alpha = alpha / 255.0
# Multiply the foreground with the alpha matte
foreground = cv2.multiply(alpha, foreground)
cv2.imwrite("images/foreground.png", foreground)
src = cv2.imread("images/foreground.png", 1)
os.remove("images/foreground.png")
tmp = cv2.cvtColor(src, cv2.COLOR_BGR2GRAY)
_, alpha = cv2.threshold(tmp, 0, 255, cv2.THRESH_BINARY)
b, g, r = cv2.split(src)
rgba = [b, g, r, alpha]
dst = cv2.merge(rgba, 4)
cv2.imwrite(path, dst)
def apply_detection_results(image,
masks,
bboxes,
class_ids,
class_names,
colors,
scores=None):
"""
Performs inference on target image and draws them on a return image.
:param image (np.ndarray): 3-channel, 8-bit RGB image. Size must be according to CONFIG file (default 640x480)
:param masks (np.ndarray): binary array of size [height, width, no_of_detections]
:param bboxes (list): list of bboxes. Each bbox is a tuple (y1, x1, y2, x2)
:param class_ids (list): one numerical ID for each detection
:param class_names (list): string of class names, as given by Dataset.class_names
:param colors (dict): keys are class names, values are float [0 : 1] 3d tuples representing RGB color
:param scores (list): list of scores, one for each detection
:return: result (image): image with detection results splashed on it, 3-channel, 8-bit RGB image
"""
# Opacity of masks: 0.5
opacity = 0.5
result = image.astype(float) / 255
for detection_idx in range(masks.shape[2]):
if not np.any(bboxes[detection_idx]):
# Skip this instance. Has no bbox. Likely lost in image cropping.
continue
# Get the color in float form
color = colors[class_names[class_ids[detection_idx]]]
# Draw the segmentation mask
mask = masks[:, :, detection_idx]
alpha_mask = np.stack((mask, mask, mask), axis=2)
alpha_mask = alpha_mask.astype(np.float) * opacity
assert alpha_mask.shape == image.shape
foreground = np.ones(image.shape, dtype=float) * color
_background = cv2.multiply(1.0 - alpha_mask, result)
_foreground = cv2.multiply(alpha_mask, foreground)
result = cv2.add(_foreground, _background)
# Draw the bounding box
y1, x1, y2, x2 = bboxes[detection_idx]
cv2.rectangle(result, (x1, y1), (x2, y2), color, thickness=1)
# Caption time
font = cv2.FONT_HERSHEY_SIMPLEX
fontScale = 0.3
lineType = 2
offset_x_text = 2
offset_y_text = -4
label = class_names[class_ids[detection_idx]]
caption = "{} {:.3f}".format(
label, scores[detection_idx]) if scores.any() else label
cv2.putText(result,
caption, (x1 + offset_x_text, y2 + offset_y_text),
fontFace=font,
fontScale=fontScale,
color=(1.0, 1.0, 1.0),
lineType=lineType)
result *= 255
result = result.astype(np.uint8)
return result
def decode_segmap(image, source, nc=21):
label_colors = np.array([
(0, 0, 0), # 0=배경
# 1=비행기, 2=자전거, 3=새, 4=배, 5=병
(128, 0, 0),
(0, 128, 0),
(128, 128, 0),
(0, 0, 128),
(128, 0, 128),
# 6=버스, 7=차, 8=고양이, 9=의자, 10=소
(0, 128, 128),
(128, 128, 128),
(64, 0, 0),
(192, 0, 0),
(64, 128, 0),
# 11=식탁, 12=개, 13=소, 14=오토바이, 15=사람
(192, 128, 0),
(64, 0, 128),
(192, 0, 128),
(64, 128, 128),
(192, 128, 128),
# 16=화분식물, 17=양, 18=소파, 19=기차, 20=TV
(0, 64, 0),
(128, 64, 0),
(0, 192, 0),
(128, 192, 0),
(0, 64, 128)
])
#r,g,b의 값에 image와 동일한 0배열 반환
r = np.zeros_like(image).astype(np.uint8)
g = np.zeros_like(image).astype(np.uint8)
b = np.zeros_like(image).astype(np.uint8)
for l in range(0, nc):
idx = image == l
r[idx] = label_colors[l, 0]
g[idx] = label_colors[l, 1]
b[idx] = label_colors[l, 2]
#추출할 객체를 인식하는 배열 합치기
rgb = np.stack([r, g, b], axis=2)
#추출할 객체 이미지 호출
foreground = cv2.imread(source)
#추출할 이미지를 rgb값으로 색깔 변환하고 사이즈 조정
foreground = cv2.cvtColor(foreground, cv2.COLOR_BGR2RGB)
foreground = cv2.resize(foreground, (r.shape[1], r.shape[0]))
#rgb 출력 맵과 크기가 같은 배경 배열 생성
background = 255 * np.ones_like(rgb).astype(np.uint8)
#uint8를 float형으로 변환
foreground = foreground.astype(float)
background = background.astype(float)
#rgb 출력 맵의 이진마스크 생성
th, alpha = cv2.threshold(np.array(rgb), 0, 255, cv2.THRESH_BINARY)
#마스크에 약간의 blur처리를 통해 가장자리 부드럽게 처리
alpha = cv2.GaussianBlur(alpha, (7, 7), 0)
#알파 마스크 정규화
alpha = alpha.astype(float) / 255
#알파 마스크와 추출할 이미지 덧붙이기
foreground = cv2.multiply(alpha, foreground)
#배경 이미지에 (1-alpha)값 덧붙이기
background = cv2.multiply(1.0 - alpha, background)
#처리된 결과물 추출
outImage = cv2.add(foreground, background)
# 이미지 저장
# plt.imsave('blog/../static/test/test{}.png'.format(i+1), outImage/255)
image_name = 'blog/../media/process/2020/11/result{}.png'.format(
str(dt.datetime.now()).replace(' ',
'').replace(':',
'_').replace('.', '_'))
plt.imsave(image_name, outImage / 255)
return image_name
# Convert uint8 to float
foreground = foreground.astype(float)
background = background.astype(float)
# Normalize the alpha mask to keep intensity between 0 and 1
alpha = alpha.astype(float) / 255
alpha_b = alpha_b.astype(float) / 255
# Multiply the foreground with the alpha matte
alpha = alpha.resize((256, 256, 3), Image.ANTIALIAS)
foreground = foreground.resize((256, 256, 3), Image.ANTIALIAS)
background = background.resize((256, 256, 3), Image.ANTIALIAS)
alpha = numpy.array(alpha)
foreground = numpy.array(foreground)
background = numpy.array(background)
alpha = alpha[:, :, ::-1].copy()
foreground = foreground[:, :, ::-1].copy()
background = background[:, :, ::-1].copy()
foreground = cv2.multiply(alpha, foreground)
# Multiply the background with ( 1 - alpha )
background = cv2.multiply(1.0 - alpha, background)
# Add the masked foreground and background.
outImage = cv2.add(foreground, background)
cv2.imwrite('out_imgi[-6:].jpg', outImage)
def mask_average(input_param_file, input_mask_file, output_dir):
try:
os.chdir(output_dir)
print(input_mask_file)
print(type(input_mask_file))
print(input_mask_file.split("\n"))
mask_files = input_mask_file.split("\n")
# print("---")
# print(mask_files[0])
# print(mask_files[1])
# print(len(mask_files))
nROI = len(mask_files)
# Load param image
param_image = nib.load(input_param_file)
size_x_param = param_image.shape[0]
size_y_param = param_image.shape[1]
try:
size_z_param = param_image.shape[2]
except:
size_z_param = None
param_hdr = param_image.header
param_qform = param_hdr['qform_code']
param_sform = param_hdr['sform_code']
canonical_param_image = nib.as_closest_canonical(param_image)
param_img_orientation = nib.aff2axcodes(canonical_param_image.affine)
param_image_data = param_image.get_fdata()
param_image_original = param_image.get_fdata()
param_filename = os.path.basename(input_param_file).split('.')[0]
# Load mask
mask_image = nib.load(mask_files[0])
mask_image.affine.shape
size_x_mask = mask_image.shape[0]
size_y_mask = mask_image.shape[1]
try:
size_z_mask = mask_image.shape[2]
except:
size_z_mask = None
mask_hdr = mask_image.header
mask_qform = mask_hdr['qform_code']
mask_sform = mask_hdr['sform_code']
canonical_mask = nib.as_closest_canonical(mask_image)
mask_img_orientation = nib.aff2axcodes(canonical_mask.affine)
mask = np.zeros((size_x_mask, size_y_mask, size_z_mask, nROI))
for i in range(nROI):
mask_image = nib.load(mask_files[i])
mask_image_data = mask_image.get_fdata()
mask[:, :, :, i] = mask_image_data
# Check image orientation
# print(param_hdr.get_sform(coded=True))
# print(param_hdr.get_qform(coded=True))
# print(mask_hdr.get_sform(coded=True))
# print(mask_hdr.get_qform(coded=True))
# mask_hdr.set_qform(param_hdr.get_sform, code='scanner')
# param_hdr.set_qform(mask_hdr.get_sform, code='scanner')
# if mask_qform == param_qform and mask_sform == param_sform:
# pass
# elif mask_qform != param_qform and mask_sform != param_sform:
# mask_hdr.set_sform(param_hdr.get_sform, code='scanner')
# affine = param_hdr.get_sform
# img = nib.load(input_mask_file)
# img.header.set_qform(affine, 1)
# img.header.set_sform(affine, 1)
# img.affine = affine
# img_data = img.get_fdata()
# plt.imshow(img_data,cmap='gray')
# img.to_filename('tryout.nii')
if size_x_mask != size_x_param and size_y_mask != size_y_param and size_z_mask != 1:
full_resized_mask = np.zeros(
(size_x_param, size_y_param, size_z_mask))
with open("statistics.txt", "w+") as f:
f.write(
"***************************************************************************** \n"
)
f.write("XNAT-PIC Pipeline: Mask Average\n")
f.write(
"Apply a mask to a parametric map and compute mean value in the ROIs\n"
)
f.write("\n")
f.write("Author: Sara Zullino \n")
f.write("Mailto: [email protected] \n")
f.write(
"***************************************************************************** \n"
)
f.write("\n")
resized_mask = np.zeros(
(size_x_param, size_y_param, size_z_mask, nROI))
for i in range(nROI):
mask_roi = mask[:, :, :, i]
#mask_copy_bin = np.sign(mask_copy)
mask_roi_resized = cv2.resize(
mask_roi,
dsize=(size_x_param, size_y_param),
interpolation=cv2.INTER_LINEAR)
mask_roi_resized_int = (mask_roi_resized != 0).astype(int)
resized_mask[:, :, :, i] = mask_roi_resized_int
temp = resized_mask[:, :, :, i]
full_resized_mask = full_resized_mask + temp
# Applying multiply method
param_image_mask = cv2.multiply(param_image_data,
resized_mask[:, :, :, i],
dtype=cv2.CV_32F)
param_image_mask_nonzero_array = param_image_mask[
param_image_mask > 0]
mean_T2 = param_image_mask_nonzero_array.mean()
std_T2 = param_image_mask_nonzero_array.std()
median_T2 = np.median(param_image_mask_nonzero_array)
print("param_image_mask_nonzero_array max is",
np.amax(param_image_mask_nonzero_array))
print("param_image_mask_nonzero_array min is",
np.amin(param_image_mask_nonzero_array))
R2_image_mask_nonzero_array = 1000 / param_image_mask_nonzero_array
mean_R2 = R2_image_mask_nonzero_array.mean()
std_R2 = R2_image_mask_nonzero_array.std()
median_R2 = np.median(R2_image_mask_nonzero_array)
# Count the number of nonzero pixels in the thresholded image
roi_area = cv2.countNonZero(resized_mask[:, :, :, i])
print("ROI number: {}".format(i + 1), file=f)
print("ROI file name: %s" %
str(os.path.basename(mask_files[i])),
file=f)
print("ROI Area: {:0.2f} ".format(roi_area), file=f)
print("Mean T2: {:0.2f} ms".format(mean_T2), file=f)
print("STD T2: {:0.2f} ms".format(std_T2, 2), file=f)
print("Median T2: {:0.2f} ms".format(median_T2), file=f)
print("Mean R2: {:0.2f} 1/s ".format(mean_R2, 2), file=f)
print("STD R2: {:0.2f} 1/s".format(std_R2, 2), file=f)
print("Median R2: {:0.2f} 1/s".format(median_R2), file=f)
f.write(
"----------------------------------------------------------------------------- \n"
)
f.close()
# Save nifti figures
param_image_full_mask = np.zeros(
(size_x_param, size_y_param, size_z_mask))
for j in range(0, size_z_mask):
param_image_full_mask[:, :,
j] = cv2.multiply(param_image_data[:, :,
j],
full_resized_mask[:, :,
j],
dtype=cv2.CV_32F)
#param_image_full_mask[:,:,j] = cv2.multiply(full_resized_mask[:,:,j], full_resized_mask[:,:,j], dtype=cv2.CV_32F)
full_resized_param_img = nib.Nifti1Image(param_image_full_mask,
param_image.affine)
full_resized_param_img.header.get_data_shape()
nib.save(full_resized_param_img, '%s_masked.nii' % param_filename)
full_resized_mask_img = nib.Nifti1Image(full_resized_mask,
mask_image.affine)
full_resized_mask_img.header.get_data_shape()
nib.save(full_resized_mask_img, 'mask.nii')
# # Figure: PARAMETRIC MAP
# def display_multiple_img(images, rows = 1, cols=1):
# figure, ax = plt.subplots(nrows=rows,ncols=cols )
# for ind,title in enumerate(images):
# ax.ravel()[ind].imshow(images[title], cmap='gray')
# ax.ravel()[ind].set_title(title)
# ax.ravel()[ind].set_axis_off()
# plt.tight_layout()
# #plt.suptitle('%s masked [s]'% param_filename)
# plt.savefig('%s_masked.png' % param_filename)
# plt.show()
# total_images = size_z_mask
# images = {str(i+1): param_image_full_mask[:,:,i] for i in range(total_images)}
# display_multiple_img(images, 1, size_z_mask)
# # Figure: MASK
# def display_multiple_img(images, rows = 1, cols=1):
# figure, ax = plt.subplots(nrows=rows,ncols=cols )
# for ind,title in enumerate(images):
# ax.ravel()[ind].imshow(images[title], cmap='gray')
# ax.ravel()[ind].set_title(title)
# ax.ravel()[ind].set_axis_off()
# plt.tight_layout()
# #plt.suptitle('mask')
# plt.savefig('mask.png')
# plt.show()
# total_images = size_z_mask
# images = {str(i+1): full_resized_mask[:,:,i] for i in range(total_images)}
# display_multiple_img(images, 1, size_z_mask)
elif size_x_mask != size_x_param and size_y_mask != size_y_param and size_z_mask == 1:
full_resized_mask = np.zeros((size_x_param, size_y_param))
if size_x_mask != size_x_param and size_y_mask != size_y_param:
resized_mask = np.zeros((size_x_param, size_y_param, nROI))
with open("statistics.txt", "w+") as f:
f.write(
"***************************************************************************** \n"
)
f.write("XNAT-PIC Pipeline: Mask Average\n")
f.write(
"Apply a mask to a parametric map and compute mean value in the ROIs\n"
)
f.write("\n")
f.write("Author: Sara Zullino \n")
f.write("Mailto: [email protected] \n")
f.write(
"***************************************************************************** \n"
)
f.write("\n")
for i in range(nROI):
mask_roi = mask[:, :, :, i]
#mask_copy_bin = np.sign(mask_copy)
mask_roi_resized = cv2.resize(
mask_roi,
dsize=(size_x_param, size_y_param),
interpolation=cv2.INTER_LINEAR)
mask_roi_resized_int = (mask_roi_resized !=
0).astype(int)
resized_mask[:, :, i] = mask_roi_resized_int
temp = resized_mask[:, :, i]
full_resized_mask = full_resized_mask + temp
# Applying multiply method
param_image_mask = cv2.multiply(param_image_data,
resized_mask[:, :, i],
dtype=cv2.CV_32F)
param_image_mask_nonzero_array = param_image_mask[
param_image_mask > 0]
mean_T2 = param_image_mask_nonzero_array.mean()
std_T2 = param_image_mask_nonzero_array.std()
median_T2 = np.median(param_image_mask_nonzero_array)
print("param_image_mask_nonzero_array max is",
np.amax(param_image_mask_nonzero_array))
print("param_image_mask_nonzero_array min is",
np.amin(param_image_mask_nonzero_array))
R2_image_mask_nonzero_array = 1000 / param_image_mask_nonzero_array
mean_R2 = R2_image_mask_nonzero_array.mean()
std_R2 = R2_image_mask_nonzero_array.std()
median_R2 = np.median(R2_image_mask_nonzero_array)
# Count the number of nonzero pixels in the thresholded image
roi_area = cv2.countNonZero(resized_mask[:, :, i])
print("ROI number: {}".format(i + 1), file=f)
print("ROI file name: %s" %
str(os.path.basename(mask_files[i])),
file=f)
print("ROI Area: {:0.2f} ".format(roi_area), file=f)
print("Mean T2: {:0.2f} ms".format(mean_T2), file=f)
print("STD T2: {:0.2f} ms".format(std_T2, 2), file=f)
print("Median T2: {:0.2f} ms".format(median_T2),
file=f)
print("Mean R2: {:0.2f} 1/s ".format(mean_R2, 2),
file=f)
print("STD R2: {:0.2f} 1/s".format(std_R2, 2), file=f)
print("Median R2: {:0.2f} 1/s".format(median_R2),
file=f)
f.write(
"----------------------------------------------------------------------------- \n"
)
f.close()
# Save nifti figures
param_image_full_mask = cv2.multiply(param_image_data,
full_resized_mask,
dtype=cv2.CV_32F)
full_resized_param_img = nib.Nifti1Image(param_image_full_mask,
param_image.affine)
full_resized_param_img.header.get_data_shape()
nib.save(full_resized_param_img, '%s_masked.nii' % param_filename)
full_resized_mask_img = nib.Nifti1Image(full_resized_mask,
mask_image.affine)
full_resized_mask_img.header.get_data_shape()
nib.save(full_resized_mask_img, 'mask.nii')
except Exception as e:
print(e)
exc_type, exc_value, exc_traceback = sys.exc_info()
traceback.print_tb(exc_traceback)
sys.exit(1)
imgCannyErosion = cv2.erode(imgCanny, kernel, iterations=1)
imgCannyDilation = cv2.dilate(imgCanny, kernel, iterations=1)
imgCannyDialThenErosion = cv2.erode(imgCannyDilation, kernel, iterations=1)
#operations
imgand = cv2.bitwise_and(img2Gray, imgGray, mask=None)
imgor = cv2.bitwise_or(img2Gray, imgGray, mask=None)
imgxor = cv2.bitwise_xor(img2Gray, imgGray, mask=None)
imgxnor = cv2.bitwise_not(imgxor, mask=None)
imgnand = cv2.bitwise_not(imgand, mask=None)
not1 = cv2.bitwise_not(imgGray, mask=None)
not2 = cv2.bitwise_not(img2Gray, mask=None)
coloradd = imgand = cv2.add(img, img2, mask=None)
colorsub = imgand = cv2.subtract(img, img2, mask=None)
colormulti = imgand = cv2.multiply(img, img2)
colordiv = imgand = cv2.divide(img, img2)
print("Image Dimension(h, w, 3): " + str(img.shape))
print("Resized Image Dimension(h, w, 3): " + str(imgResize.shape))
print("Cropped Image Dimension(h, w, 3): " + str(imgCropped.shape))
cv2.imwrite("Gray.jpg", imgGray)
cv2.imwrite("Blur.jpg", imgBlur)
cv2.imwrite("Canny.jpg", imgCanny)
cv2.imwrite("Dilation.jpg", imgDilation)
cv2.imwrite("Erosion.jpg", imgErosion)
cv2.imwrite("Resize.jpg", imgResize)
cv2.imwrite("Cropped.jpg", imgCropped)
cv2.imwrite("CannyDilation.jpg", imgCannyDilation)
cv2.imwrite("CannyErosion.jpg", imgCannyErosion)
if (np.sum(I) == 0):
mean_sub[counter_row, counter_col] = 0
std_sub[counter_row, counter_col] = 1
else:
means = np.mean(temp)
stddevs = np.sqrt(np.var(temp)) #Computation of mean and std
mean_sub[counter_row, counter_col] = means
std_sub[counter_row, counter_col] = stddevs
row_img, col_img = img.shape
mean_full = np.zeros((row_img, col_img), dtype=np.double)
cv2.resize(mean_sub, (col_img, row_img),
mean_full,
interpolation=cv2.INTER_CUBIC) # Interpolate the mean_dev
mask = mask.astype(float)
cv2.multiply(mean_full, mask, mean_full) #multiplication by mask
std_full = cv2.resize(std_sub, (col_img, row_img)) #interpolate stddev
cv2.multiply(std_full, mask, std_full) #multiplication by mask
img_sub = cv2.subtract(img, mean_full)
pcm_dist = cv2.divide(img_sub, std_full)
pcm_dist = np.abs(
pcm_dist
) # Compute the absolute value of the distance as it should be positive
cv2.multiply(pcm_dist, mask, pcm_dist)
pcm_dist[pcm_dist < 1] = 1 #choose the threshold globally as 1
pcm_dist[pcm_dist != 1] = 0
block_r = 512
block_c = 512
for counter_row in range(sz_r / block_r):
height, width = img_color.shape[:2]
img_color = cv.resize(img_color, (width, height), interpolation=cv.INTER_AREA)
# 원본 영상을 HSV 영상으로 변환합니다.
img_hsv = cv.cvtColor(img_color, cv.COLOR_BGR2HSV)
# 범위 값으로 HSV 이미지에서 마스크를 생성합니다.
img_mask1 = cv.inRange(img_hsv, lower_blue1, upper_blue1)
img_mask2 = cv.inRange(img_hsv, lower_blue2, upper_blue2)
img_mask3 = cv.inRange(img_hsv, lower_blue3, upper_blue3)
img_mask = img_mask1 | img_mask2 | img_mask3
img_mask = cv.bitwise_not(img_mask)
blurred = cv.GaussianBlur(img_mask, (11, 11), 0)
blurred1 = cv.cvtColor(blurred, cv.COLOR_GRAY2BGR)
blurred1 = blurred1.astype(float)/255
img_color1 = img_color.astype(float)
img_result = cv.multiply(blurred1, img_color1)
cv.imshow('img_color', img_color)
cv.imshow('img_mask', img_mask)
img_result = np.uint8(img_result)
cv.imshow('img_result', img_result)
# ESC 키누르면 종료
if cv.waitKey(1) & 0xFF == 27:
break
cv.destroyAllWindows()
# D = avg(Xpow) -----------------------------------------------------------------------------------------------
cv2.reduce(matXpow, 1, cv2.cv.CV_REDUCE_AVG, vetD, -1)
# Y = inv(D)*X-------------------------------------------------------------------------------------------------
for i in xrange(0, size_img):
matY[i, :, 0] = (1 / vetD[i]) * matX[i, :, 0]
for i in xrange(0, size_img):
matY[i, :, 1] = (1 / vetD[i]) * matX[i, :, 1]
# X = conjugate(X)--------------------------------------------------------------------------------------------
for i in xrange(0, size_img):
cv2.multiply(matX[i, :, 1], np.ones((total_img, 1)), matImX[i, :], -1,
-1)
matX[:, :, 1] = matImX
# print("--- X --- size: {} ".format(matX.shape))
# print matX[:, :, 1]
# X'-----------------------------------------------------------
matXtr[:, :, 0] = cv2.transpose(matX[:, :, 0])
matXtr[:, :, 1] = cv2.transpose(matX[:, :, 1])
# print("\n\n--- X' --- size: {} ".format(matXtr.shape))
# print matXtr
# Z = X'*Y ----------------------------------------------------
np.uint8(gaussian_star_finder.mask))
result = image.copy()
cv2.drawKeypoints(result, kp, result, (0, 0, 255), 1)
else:
log_star_detector.kernel_size = kernel_size
log_star_detector.block_size = block_size
log_star_detector.threshold = threshold
candidate_finder.setImage(image)
if image_switch == 2:
result = log_star_detector.debug
masked = cv2.multiply(result, 1 - saturated)
result = cv2.multiply(masked, gaussian_star_finder.mask) * 255
else:
result = image.copy()
candidate_finder.drawCandidates(result)
cv2.imshow(window, result)
k = cv2.waitKey(30) & 0xFF
if k == 27:
break
if k == ord('s'):
filename = 'out.png'
print('Saving ' + filename)
cv2.imwrite(filename, result)
import cv2
import numpy as np
img = cv2.imread("lena.jpg")
cv2.imshow("Origianl image", img)
cv2.waitKey(0)
#M=np.ones(img.shape,dtype="uint8")*150
M = np.zeros(img.shape, dtype="uint8") + 150
added = cv2.add(img, M)
cv2.imshow("Added", added)
cv2.waitKey(0)
subtracted = cv2.subtract(img, M)
cv2.imshow("Subtract", subtracted)
cv2.waitKey(0)
mul = cv2.multiply(img, M)
cv2.imshow("Multiply", mul)
cv2.waitKey(0)
cv2.destroyAllWindows()
def __drawDNA(self, DNA, inImg):
#get DNA data
color = DNA[0]
posX = int(
DNA[2]) + self.padding #add padding since indices have shifted
posY = int(DNA[1]) + self.padding
size = DNA[3]
rotation = DNA[4]
brushNumber = int(DNA[5])
#load brush alpha
brushImg = self.brushes[brushNumber]
#resize the brush
brushImg = cv2.resize(brushImg,
None,
fx=size,
fy=size,
interpolation=cv2.INTER_CUBIC)
#rotate
brushImg = self.__rotateImg(brushImg, rotation)
#brush img data
brushImg = cv2.cvtColor(brushImg, cv2.COLOR_BGR2GRAY)
rows, cols = brushImg.shape
#create a colored canvas
myClr = np.copy(brushImg)
myClr[:, :] = color
#find ROI
inImg_rows, inImg_cols = inImg.shape
y_min = int(posY - rows / 2)
y_max = int(posY + (rows - rows / 2))
x_min = int(posX - cols / 2)
x_max = int(posX + (cols - cols / 2))
# Convert uint8 to float
foreground = myClr[0:rows, 0:cols].astype(float)
background = inImg[y_min:y_max, x_min:x_max].astype(float) #get ROI
# Normalize the alpha mask to keep intensity between 0 and 1
alpha = brushImg.astype(float) / 255.0
try:
# Multiply the foreground with the alpha matte
foreground = cv2.multiply(alpha, foreground)
# Multiply the background with ( 1 - alpha )
background = cv2.multiply(np.clip((1.0 - alpha), 0.0, 1.0),
background)
# Add the masked foreground and background.
outImage = (np.clip(cv2.add(foreground, background), 0.0,
255.0)).astype(np.uint8)
inImg[y_min:y_max, x_min:x_max] = outImage
except:
print('------ \n', 'in image ', inImg.shape)
print('pivot: ', posY, posX)
print('brush size: ', self.brushSide)
print('brush shape: ', brushImg.shape)
print(" Y range: ", rangeY, 'X range: ', rangeX)
print('bg coord: ', posY, posY + rangeY, posX, posX + rangeX)
print('fg: ', foreground.shape)
print('bg: ', background.shape)
print('alpha: ', alpha.shape)
return inImg
import cv2
import numpy as np
ksize_morph = np.ones((7, 7), np.uint8)
ksize_filter = (15, 15)
sigmaX = 2.4
sigmaY = 2.4
filename = '/home/yashmanian/images/Imgs/ironman.jpg'
img = cv2.imread(filename)
height, width, channels = img.shape
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray, (11, 11), 2, 2, cv2.BORDER_CONSTANT)
ret, bin = cv2.threshold(gray, 180, 255, cv2.THRESH_BINARY)
masked = cv2.multiply(gray, bin)
masked = np.float32(masked)
dst = cv2.cornerHarris(masked, 5, 7, 0.04)
dst = cv2.dilate(dst, ksize_morph, iterations=1)
cv2.imshow(dst)
if cv2.waitKey(0) & 0xff == 27:
cv2.destroyAllWindows()
background = cv2.imread("radar_bg.png", -1)
alpha = foreground[..., 3]
# alpha = cv2.imread("puppets_alpha.png")
# Convert uint8 to float
foreground = foreground.astype(float)
background = background.astype(float)
# Normalize the alpha mask to keep intensity between 0 and 1
alpha = alpha.astype(float) / 255
t1 = time.time()
for _ in range(times):
# Multiply the foreground with the alpha matte
foreground[..., 0] = cv2.multiply(alpha, foreground[..., 0])
foreground[..., 1] = cv2.multiply(alpha, foreground[..., 1])
foreground[..., 2] = cv2.multiply(alpha, foreground[..., 2])
# Multiply the background with ( 1 - alpha )
background[..., 0] = cv2.multiply(1.0 - alpha, background[..., 0])
background[..., 1] = cv2.multiply(1.0 - alpha, background[..., 1])
background[..., 2] = cv2.multiply(1.0 - alpha, background[..., 2])
# Add the masked foreground and background.
outImage = cv2.add(foreground, background)
t2 = time.time()
print(1e3 * (t2 - t1))
# # Display image
def analyze_object(img,
imgname,
obj,
mask,
device,
debug=False,
filename=False):
# Outputs numeric properties for an input object (contour or grouped contours)
# Also color classification?
# img = image object (most likely the original), color(RGB)
# imgname= name of image
# obj = single or grouped contour object
# device = device number. Used to count steps in the pipeline
# debug= True/False. If True, print image
# filename= False or image name. If defined print image
device += 1
ori_img = np.copy(img)
if len(np.shape(img)) == 3:
ix, iy, iz = np.shape(img)
else:
ix, iy = np.shape(img)
size = ix, iy, 3
size1 = ix, iy
background = np.zeros(size, dtype=np.uint8)
background1 = np.zeros(size1, dtype=np.uint8)
background2 = np.zeros(size1, dtype=np.uint8)
# Check is object is touching image boundaries (QC)
frame_background = np.zeros(size1, dtype=np.uint8)
frame = frame_background + 1
frame_contour, frame_heirarchy = cv2.findContours(frame, cv2.RETR_TREE,
cv2.CHAIN_APPROX_NONE)
ptest = []
vobj = np.vstack(obj)
for i, c in enumerate(vobj):
xy = tuple(c)
pptest = cv2.pointPolygonTest(frame_contour[0], xy, measureDist=False)
ptest.append(pptest)
in_bounds = all(c == 1 for c in ptest)
# Convex Hull
hull = cv2.convexHull(obj)
hull_vertices = len(hull)
# Moments
# m = cv2.moments(obj)
m = cv2.moments(mask, binaryImage=True)
## Properties
# Area
area = m['m00']
if area:
# Convex Hull area
hull_area = cv2.contourArea(hull)
# Solidity
solidity = area / hull_area
# Perimeter
perimeter = cv2.arcLength(obj, closed=True)
# x and y position (bottom left?) and extent x (width) and extent y (height)
x, y, width, height = cv2.boundingRect(obj)
# Centroid (center of mass x, center of mass y)
cmx, cmy = (m['m10'] / m['m00'], m['m01'] / m['m00'])
# Ellipse
center, axes, angle = cv2.fitEllipse(obj)
major_axis = np.argmax(axes)
minor_axis = 1 - major_axis
major_axis_length = axes[major_axis]
minor_axis_length = axes[minor_axis]
eccentricity = np.sqrt(1 - (axes[minor_axis] / axes[major_axis])**2)
#Longest Axis: line through center of mass and point on the convex hull that is furthest away
cv2.circle(background, (int(cmx), int(cmy)), 4, (255, 255, 255), -1)
center_p = cv2.cvtColor(background, cv2.COLOR_BGR2GRAY)
ret, centerp_binary = cv2.threshold(center_p, 0, 255,
cv2.THRESH_BINARY)
centerpoint, cpoint_h = cv2.findContours(centerp_binary, cv2.RETR_TREE,
cv2.CHAIN_APPROX_NONE)
dist = []
vhull = np.vstack(hull)
for i, c in enumerate(vhull):
xy = tuple(c)
pptest = cv2.pointPolygonTest(centerpoint[0], xy, measureDist=True)
dist.append(pptest)
abs_dist = np.absolute(dist)
max_i = np.argmax(abs_dist)
caliper_max_x, caliper_max_y = list(tuple(vhull[max_i]))
caliper_mid_x, caliper_mid_y = [int(cmx), int(cmy)]
xdiff = float(caliper_max_x - caliper_mid_x)
ydiff = float(caliper_max_y - caliper_mid_y)
if xdiff != 0:
slope = (float(ydiff / xdiff))
if xdiff == 0:
slope = 1
b_line = caliper_mid_y - (slope * caliper_mid_x)
if slope == 0:
xintercept = 0
xintercept1 = 0
yintercept = 'none'
yintercept1 = 'none'
cv2.line(background1, (iy, caliper_mid_y), (0, caliper_mid_y),
(255), 1)
else:
xintercept = int(-b_line / slope)
xintercept1 = int((ix - b_line) / slope)
yintercept = 'none'
yintercept1 = 'none'
if 0 <= xintercept <= iy and 0 <= xintercept1 <= iy:
cv2.line(background1, (xintercept1, ix), (xintercept, 0),
(255), 1)
elif xintercept < 0 or xintercept > iy or xintercept1 < 0 or xintercept1 > iy:
if xintercept < 0 and 0 <= xintercept1 <= iy:
yintercept = int(b_line)
cv2.line(background1, (0, yintercept), (xintercept1, ix),
(255), 1)
elif xintercept > iy and 0 <= xintercept1 <= iy:
yintercept1 = int((slope * iy) + b_line)
cv2.line(background1, (iy, yintercept1), (xintercept1, ix),
(255), 1)
elif 0 <= xintercept <= iy and xintercept1 < 0:
yintercept = int(b_line)
cv2.line(background1, (0, yintercept), (xintercept, 0),
(255), 1)
elif 0 <= xintercept <= iy and xintercept1 > iy:
yintercept1 = int((slope * iy) + b_line)
cv2.line(background1, (iy, yintercept1), (xintercept, 0),
(255), 1)
else:
yintercept = int(b_line)
yintercept1 = int((slope * iy) + b_line)
cv2.line(background1, (0, yintercept), (iy, yintercept1),
(255), 1)
ret1, line_binary = cv2.threshold(background1, 0, 255,
cv2.THRESH_BINARY)
#print_image(line_binary,(str(device)+'_caliperfit.png'))
cv2.drawContours(background2, [hull], -1, (255), -1)
ret2, hullp_binary = cv2.threshold(background2, 0, 255,
cv2.THRESH_BINARY)
#print_image(hullp_binary,(str(device)+'_hull.png'))
caliper = cv2.multiply(line_binary, hullp_binary)
#print_image(caliper,(str(device)+'_caliperlength.png'))
caliper_y, caliper_x = np.array(caliper.nonzero())
caliper_matrix = np.vstack((caliper_x, caliper_y))
caliper_transpose = np.transpose(caliper_matrix)
caliper_length = len(caliper_transpose)
caliper_transpose1 = np.lexsort((caliper_y, caliper_x))
caliper_transpose2 = [(caliper_x[i], caliper_y[i])
for i in caliper_transpose1]
caliper_transpose = np.array(caliper_transpose2)
else:
hull_area, solidity, perimeter, width, height, cmx, cmy = 'ND', 'ND', 'ND', 'ND', 'ND', 'ND', 'ND'
#Store Shape Data
shape_header = ('HEADER_SHAPES', 'area', 'hull-area', 'solidity',
'perimeter', 'width', 'height', 'longest_axis',
'center-of-mass-x', 'center-of-mass-y', 'hull_vertices',
'in_bounds')
shape_data = ('SHAPES_DATA', area, hull_area, solidity, perimeter, width,
height, caliper_length, cmx, cmy, hull_vertices, in_bounds)
#Draw properties
if area and filename:
cv2.drawContours(ori_img, obj, -1, (255, 0, 0), 1)
cv2.drawContours(ori_img, [hull], -1, (0, 0, 255), 1)
cv2.line(ori_img, (x, y), (x + width, y), (0, 0, 255), 1)
cv2.line(ori_img, (int(cmx), y), (int(cmx), y + height), (0, 0, 255),
1)
cv2.line(ori_img, (tuple(caliper_transpose[caliper_length - 1])),
(tuple(caliper_transpose[0])), (0, 0, 255), 1)
cv2.circle(ori_img, (int(cmx), int(cmy)), 10, (0, 0, 255), 1)
# Output images with convex hull, extent x and y
extention = filename.split('.')[-1]
out_file = str(filename[0:-4]) + '_shapes.' + extention
print_image(ori_img, out_file)
print('\t'.join(map(str, ('IMAGE', 'shapes', out_file))))
else:
pass
if debug:
cv2.drawContours(ori_img, obj, -1, (255, 0, 0), 1)
cv2.drawContours(ori_img, [hull], -1, (0, 0, 255), 1)
cv2.line(ori_img, (x, y), (x + width, y), (0, 0, 255), 1)
cv2.line(ori_img, (int(cmx), y), (int(cmx), y + height), (0, 0, 255),
1)
cv2.circle(ori_img, (int(cmx), int(cmy)), 10, (0, 0, 255), 1)
cv2.line(ori_img, (tuple(caliper_transpose[caliper_length - 1])),
(tuple(caliper_transpose[0])), (0, 0, 255), 1)
print_image(ori_img, (str(device) + '_shapes.png'))
return device, shape_header, shape_data, ori_img
def letsHat(img, hat):
# Get alpha channel for the hat img
r, g, b, a = cv2.split(hat_im)
hat_rgb = cv2.merge((r, g, b))
cv2.imwrite("hat_alpha.jpg", a)
cv2.imwrite("hat_rgb.jpg", hat_rgb)
# Trained dlib face key points detector
predictor_path = "shape_predictor_5_face_landmarks.dat"
predictor = dlib.shape_predictor(predictor_path)
# Dlib front face detector
detector = dlib.get_frontal_face_detector()
# Front face detection result
face_detect = detector(img, 1)
# If face detected
if len(face_detect) > 0:
for d in face_detect: # for each face detected
x, y, w, h = d.left(), d.top(), (d.right() -
d.left()), (d.bottom() - d.top())
# imgRect = cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2, 8, 0)
# 5 key points detection
shape = predictor(img, d)
# for point in shape.parts():
# face_pts = cv2.circle(img, (point.x, point.y), 3, color = (0, 255, 0))
# Draw 5 feature pts one by one
# cv2.imshow('image', face_pts)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
## Draw detection retangle and pts on face
# cv2.imshow('image', img)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# Select outermost feature pts on left and right eyes
pt1 = shape.part(0)
pt2 = shape.part(2)
# print pt1, pt2
# Calculate centre pt of eye
centre_pt = ((pt1.x + pt2.x) // 2, (pt1.y + pt2.y) // 2)
# face_centrept = cv2.circle(img, centre_pt, 3, color = (0, 255, 0))
## Draw centre pts on face
# cv2.imshow('image', img)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# Adjust hat size according to face size
# # shape[0]=width, shape[1]=height
# print img.shape[0], img.shape[1]
factor = 1.5
resize_hat_h = int(
round(hat_rgb.shape[0] * w / hat_rgb.shape[1] * factor))
resize_hat_w = int(
round(hat_rgb.shape[1] * w / hat_rgb.shape[1] * factor))
if resize_hat_h > y:
resize_hat_h = y - 1
resize_hat = cv2.resize(hat_rgb, (resize_hat_w, resize_hat_h))
# cv2.imshow('image', resize_hat)
# cv2.imshow('image2', hat_rgb)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# Make resize mask from alpha channel
mask = cv2.resize(a, (resize_hat_w, resize_hat_h))
mask_inv = cv2.bitwise_not(mask)
# Hat skew wrt face detection rectangle
dh = 0
dw = 0
# ROI of figure image
roi = img[(y + dh - resize_hat_h):(y + dh),
(x + dw):(x + resize_hat_w + dw)]
# imgRect = cv2.rectangle(img, (x + dw, y + dh - resize_hat_h), (x + resize_hat_w + dw, y + dh), (255, 0, 0), 2, 8, 0)
# imgRect = cv2.rectangle(img, (x + dw, y + dh), (x + dw, y + dh), (0, 2, 0), 2, 8, 0)
# cv2.imshow('image', img)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
roi = img[(y + dh - resize_hat_h) : (y + dh), \
(centre_pt[0] - resize_hat_w // 3) : (centre_pt[0] + resize_hat_w // 3 * 2)]
# Extract hat space in ROI
roi = roi.astype(float)
# print mask_inv
mask_inv = cv2.merge((mask_inv, mask_inv, mask_inv))
alpha = mask_inv.astype(float) / 255
# print alpha
if alpha.shape != roi.shape:
alpha = cv2.resize(alpha, (roi.shape[1], roi.shape[0]))
bg = cv2.multiply(alpha, roi)
bg = bg.astype('uint8')
# cv2.imshow('imge', bg)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# Extract hat region
hat_region = cv2.bitwise_and(resize_hat, resize_hat, mask=mask)
cv2.imwrite('hat.jpg', hat_region)
# cv2.imshow('image', hat_region)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# print bg.shape, hat_region.shape
if bg.shape != hat_region.shape:
hat_region = cv2.resize(hat_region, (bg.shape[1], bg.shape[0]))
# Add the two ROI (add hat to background image)
add_hat = cv2.add(bg, hat_region)
# cv2.imshow('addhat',add_hat)
# cv2.imshow('hat', hat_region)
# cv2.imshow('bg', bg)
# cv2.imshow('original', img)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# Put the hat added region back to original image
img[(y + dh - resize_hat_h) : (y + dh), \
(centre_pt[0] - resize_hat_w // 3) : (centre_pt[0] + resize_hat_w // 3 * 2)]\
= add_hat
# Show the result and save
cv2.imshow('original', img)
cv2.waitKey(0)
cv2.destroyAllWindows()
return img
else:
print "No Face Detected!!!"
def multiply_demo(m1, m2):
# 相乘
dst = cv.multiply(m1, m2)
cv.imshow("multiply_demo", dst)
import numpy as np
import cv2
import cv2.cv as cv
from PIL import Image
import scipy
import scipy.io
from matplotlib import pyplot as plt
from computebgimg import computebgimg
from LumConDrift import LumConDrift
image = cv2.imread(
'C:\Users\AK PUJITHA\Desktop\iiit h\semester 6\honors project 2\Image Enahancement-Matlab\output\image010.png',
1)
hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
bgImg, fundusMask = computebgimg(image)
bgImg = cv2.multiply(image[:, :, 1].astype(float), bgImg.astype(float))
ldrift, cdrift = LumConDrift(bgImg, fundusMask)
g = image[:, :, 1].astype(float)
imgCorr = cv2.divide(cv2.subtract(g, ldrift), (cv2.add(cdrift, 0.0001)))
imgCorr = cv2.multiply(imgCorr, fundusMask.astype(float))
imgCorr = cv2.add(imgCorr, np.abs(np.min(imgCorr)))
imgCorr = cv2.divide(imgCorr, np.max(imgCorr))
imgCorr = cv2.multiply(imgCorr, fundusMask.astype(float))
image = image.astype(float)
image[:, :, 0] = cv2.divide(cv2.multiply(imgCorr, image[:, :, 0]),
hsv[:, :, 2].astype(float))
image[:, :, 1] = cv2.divide(cv2.multiply(imgCorr, image[:, :, 1]),
##nose=nose[0]
pts.append(chin[0])
pts.append(chin[16])
pts.append(nose[2])
pts=np.float32(pts)
## nosepts=(np.array(nosepts).reshape((-1,1,2)))
##for cnt in nosepts:
## cv2.draw(frame, [cnt], -1, (0,0,255), -1)
glasstemp=transform(pts, glass_idx)
alpha=glasstemp[:,:,3]/255.0
glasstemp=glasstemp[:,:,0:3]
for c in range(3):
frame[:,:,c]=cv2.multiply(1-alpha,frame[:,:,c])
for c in range(3):
glasstemp[:,:,c]=cv2.multiply(alpha, glasstemp[:,:,c])
frame=cv2.add(frame, glasstemp)
'''
## for drawing outlines of the eyes
eye1=cv2.convexHull(np.array(lefteye))
eye2=cv2.convexHull(np.array(righteye))
cv2.drawContours(frame, [eye1], -1, (0, 255, 0), 1)
cv2.drawContours(frame, [eye2], -1, (0, 255, 0), 1)
'''
import os
import cv2
import matplotlib.pyplot as plt
path = "Images/"
image1 = cv2.imread(path + "image.jpg", 1) # 1 for RGB, 0 for B&W
image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
image2 = cv2.imread(path + "lena.jpg", 1)
image2 = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
add = cv2.add(image2, image1)
sub = cv2.subtract(image2, image1)
mul = cv2.multiply(image2, image1)
div = cv2.divide(image2, image1)
plt.subplot(2, 3, 1)
plt.imshow(image1)
plt.title("First Image")
plt.subplot(2, 3, 2)
plt.imshow(image2)
plt.title("Second Image")
plt.subplot(2, 3, 3)
plt.imshow(add)
plt.title("Addition")
plt.subplot(2, 3, 4)
plt.imshow(sub)
cv2.destroyAllWindows()
out_vid = []
trf = T.Compose([T.Resize(640),T.ToTensor(),T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) ])
for i in range(len(frames)):
ip = trf(frames[i]).unsqueeze(0)
op = dlab(ip)['out']
om = torch.argmax(op.squeeze(), dim=0).detach().cpu().numpy()
print(om.shape)
rgb = decode_segmap(om)
output = rgb*frame_np[i]
out_vid.append(output)
final_img = []
back = np.array(Image.open("./background.jpg").resize((640,640), Image.BILINEAR), np.float64)/255
for i in range(len(frames)):
temp = out_vid[i].astype(float)
thr, alpha = cv2.threshold(temp, 0, 255, cv2.THRESH_BINARY)
alpha = cv2.GaussianBlur(alpha, (7,7), 0)
alpha = alpha/255
fore = cv2.multiply(alpha,temp)
backg = cv2.multiply((1-alpha),back)
outp = cv2.add(fore,backg)
final_img.append(255*outp)
out = cv2.VideoWriter("./output.avi",cv2.VideoWriter_fourcc('M','P','E','G'), 30, (640,640))
for i in range(len(final_img)):
out.write(np.uint8(final_img[i]))
out.release()
def rechteck():
schwarz = False
begrenzung = []
#img = fotoMachen()
# Farbkonvertierung
# hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
# Video in drei Farbkanaele splitten
# h, s, v= cv2.split(hsv)
b, g, r = cv2.split(img)
# masken berechnen
satu = 30
vis = 20
# s_mask = cv2.inRange(s, (44*2.55), (77*2.55))
# v_mask = cv2.inRange(v, 0, (13.7*2.55))
b_mask = cv2.inRange(b, 0, 45)
g_mask = cv2.inRange(g, 0, 25)
r_mask = cv2.inRange(r, 0, 45)
# Multiplizieren der Einzelmasken
mask = cv2.multiply(g_mask, r_mask)
mask = cv2.multiply(mask, b_mask)
#Dilation-Maske
kernel = np.ones((2, 2), np.uint8)
mask = cv2.dilate(mask, kernel, iterations=2)
#Areas finden
contours, hierarchy = cv2.findContours(mask, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
maxArea = 0
#groesste schwarze Flaeche
for index in range(len(contours)):
area = cv2.contourArea(contours[index])
if area > maxArea:
maxArea = area
i = index
schwarz = True
#Diese kennzeichnen
if (schwarz):
x1, y1, w1, h1 = cv2.boundingRect(contours[i])
cv2.rectangle(mask, (x1 - 10, y1 - 10), (x1 + w1, y1 + h1),
(100, 255, 100), 3)
vergleichsH = h1 * 0.9
else:
print("Etwas ist schief gelaufen, es gibt keine schwarze Fläche")
cv2.imshow("Schief gelaufen", mask)
sendNoteOn(6, 0)
return
#Postition der schwarzen Steine im Array "contours" umranden
maxArea1 = 0
for index in range(len(contours)):
area = cv2.contourArea(contours[index])
x, y, w, h = cv2.boundingRect(contours[index])
if h > vergleichsH:
print("Es gibt eine Flaeche, die ins Raster faellt")
begrenzung.append(index)
cv2.rectangle(mask, (x1 - 10, y1 - 10), (x1 + w1, y1 + h1),
(100, 255, 100), 3)
i1 = index
if h != h1:
print("2. Fläche gefunden")
x2 = x
y2 = y
w2 = w
h2 = h
maxArea1 = area
gefundeneSteine = len(begrenzung)
cv2.imshow('schwarze Steine', mask)
#falls keine Hoehe vergleichbar ist, zweit groesste Flaeche suchen
if gefundeneSteine < 2:
print("Es gibt nur 1 Flaeche, die ins Raster gefallen ist")
for index in range(len(contours)):
area = cv2.contourArea(contours[index])
x, y, w, h = cv2.boundingRect(contours[index])
if (area > maxArea1 and area != maxArea):
maxArea1 = area
i1 = index
x2, y2, w2, h2 = cv2.boundingRect(contours[i1])
cv2.rectangle(mask, (x2 - 10, y2 - 10), (x2 + w2, y2 + h2),
(100, 255, 100), 3)
if (y2 + h2 - 100) < (y2 + h2) < (y2 + h2 + 100):
begrenzung.append(i1)
cv2.imshow('schwarze Steine', mask)
print(h1, vergleichsH, h2)
print(begrenzung)
#begrenzung.append(i1)
#print (maxArea, maxArea1)
if (len(begrenzung) != 2):
print(
"Etwas ist schief gelaufen, es gibt nur schwarze Fläche oder mehr als 2"
)
cv2.imshow("Schief gelaufen", mask)
sendNoteOn(6, 0)
return
position = [x1, y1, w1, h1, x2, y2, w2, h2, maxArea1]
return (position)
def add_hat(img, hat_img):
# 分离rgba通道，合成rgb三通道帽子图，a通道后面做mask用
r, g, b, a = cv2.split(hat_img)
rgb_hat = cv2.merge((r, g, b))
cv2.imwrite("hat_alpha.jpg", a)
# ------------------------- 用dlib的人脸检测代替OpenCV的人脸检测-----------------------
# # 灰度变换
# gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
# # 用opencv自带的人脸检测器检测人脸
# face_cascade = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")
# faces = face_cascade.detectMultiScale(gray,1.05,3,cv2.CASCADE_SCALE_IMAGE,(50,50))
# ------------------------- 用dlib的人脸检测代替OpenCV的人脸检测-----------------------
# dlib人脸关键点检测器
predictor_path = "shape_predictor_5_face_landmarks.dat"
predictor = dlib.shape_predictor(predictor_path)
# dlib正脸检测器
detector = dlib.get_frontal_face_detector()
# 正脸检测
dets = detector(img, 1)
# 如果检测到人脸
if len(dets) > 0:
for d in dets:
x, y, w, h = d.left(), d.top(
), d.right() - d.left(), d.bottom() - d.top()
# x,y,w,h = faceRect
# cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2,8,0)
# 关键点检测，5个关键点
shape = predictor(img, d)
# for point in shape.parts():
# cv2.circle(img,(point.x,point.y),3,color=(0,255,0))
# cv2.imshow("image",img)
# cv2.waitKey()
# 选取左右眼眼角的点
point1 = shape.part(0)
point2 = shape.part(2)
# 求两点中心
eyes_center = ((point1.x + point2.x) // 2,
(point1.y + point2.y) // 2)
# cv2.circle(img,eyes_center,3,color=(0,255,0))
# cv2.imshow("image",img)
# cv2.waitKey()
# 根据人脸大小调整帽子大小
factor = 1.5
resized_hat_h = int(
round(rgb_hat.shape[0] * w / rgb_hat.shape[1] * factor))
resized_hat_w = int(
round(rgb_hat.shape[1] * w / rgb_hat.shape[1] * factor))
if resized_hat_h > y:
resized_hat_h = y - 1
# 根据人脸大小调整帽子大小
resized_hat = cv2.resize(rgb_hat, (resized_hat_w, resized_hat_h))
# 用alpha通道作为mask
mask = cv2.resize(a, (resized_hat_w, resized_hat_h))
mask_inv = cv2.bitwise_not(mask)
# 帽子相对与人脸框上线的偏移量
dh = 0
dw = 0
# 原图ROI
# bg_roi = img[y+dh-resized_hat_h:y+dh, x+dw:x+dw+resized_hat_w]
bg_roi = img[y + dh - resized_hat_h:y + dh,
(eyes_center[0] -
resized_hat_w // 3):(eyes_center[0] +
resized_hat_w // 3 * 2)]
# 原图ROI中提取放帽子的区域
bg_roi = bg_roi.astype(float)
mask_inv = cv2.merge((mask_inv, mask_inv, mask_inv))
alpha = mask_inv.astype(float) / 255
# 相乘之前保证两者大小一致（可能会由于四舍五入原因不一致）
alpha = cv2.resize(alpha, (bg_roi.shape[1], bg_roi.shape[0]))
# print("alpha size: ",alpha.shape)
# print("bg_roi size: ",bg_roi.shape)
bg = cv2.multiply(alpha, bg_roi)
bg = bg.astype('uint8')
cv2.imwrite("bg.jpg", bg)
# cv2.imshow("image",img)
# cv2.waitKey()
# 提取帽子区域
hat = cv2.bitwise_and(resized_hat, resized_hat, mask=mask)
cv2.imwrite("hat.jpg", hat)
# cv2.imshow("hat",hat)
# cv2.imshow("bg",bg)
# print("bg size: ",bg.shape)
# print("hat size: ",hat.shape)
# 相加之前保证两者大小一致（可能会由于四舍五入原因不一致）
hat = cv2.resize(hat, (bg_roi.shape[1], bg_roi.shape[0]))
# 两个ROI区域相加
add_hat = cv2.add(bg, hat)
# cv2.imshow("add_hat",add_hat)
# 把添加好帽子的区域放回原图
img[y + dh - resized_hat_h:y + dh,
(eyes_center[0] -
resized_hat_w // 3):(eyes_center[0] +
resized_hat_w // 3 * 2)] = add_hat
# 展示效果
# cv2.imshow("img",img )
# cv2.waitKey(0)
return img
import cv2
import numpy as np
img=cv2.imread('C:/Users/yogesh yadav/Desktop/lena.jpg')
cv2.imshow('LENA_original',img)
cv2.waitKey(0)
a=np.ones(img.shape,dtype="uint8")*100
added=cv2.add(img,a)
cv2.imshow("added",added)
cv2.waitKey(0)
multi=cv2.multiply(img,a)
cv2.imshow("muultiply",multi)
cv2.waitKey(0)
cv2.destroyAllWindows()
def addHat(img, hat_img):
# 分离rgba通道，合成rgb三通道帽子图，a通道后面做mask用
r, g, b, a = cv2.split(hat_img)
rgbHat = cv2.merge((r, g, b))
# dlib人脸关键点检测器,正脸检测
dets = detector(img, 1)
# 如果检测到人脸
if len(dets) > 0:
for d in dets:
x, y, w, h = d.left(), d.top(
), d.right() - d.left(), d.bottom() - d.top()
# 关键点检测，5个关键点")
shape = predictor(img, d)
# 选取左右眼眼角的点")
point1 = shape.part(0)
point2 = shape.part(2)
# 求两点中心
eyes_center = ((point1.x + point2.x) // 2,
(point1.y + point2.y) // 2)
# 根据人脸大小调整帽子大小
factor = 1.5
resizedHatH = int(
round(rgbHat.shape[0] * w / rgbHat.shape[1] * factor))
resizedHatW = int(
round(rgbHat.shape[1] * w / rgbHat.shape[1] * factor))
if resizedHatH > y:
resizedHatH = y - 1
# 根据人脸大小调整帽子大小
resizedHat = cv2.resize(rgbHat, (resizedHatW, resizedHatH))
# 用alpha通道作为mask
mask = cv2.resize(a, (resizedHatW, resizedHatH))
maskInv = cv2.bitwise_not(mask)
# 帽子相对与人脸框上线的偏移量
dh = 0
bgRoi = img[y + dh - resizedHatH:y + dh,
(eyes_center[0] -
resizedHatW // 3):(eyes_center[0] +
resizedHatW // 3 * 2)]
# 原图ROI中提取放帽子的区域
bgRoi = bgRoi.astype(float)
maskInv = cv2.merge((maskInv, maskInv, maskInv))
alpha = maskInv.astype(float) / 255
# 相乘之前保证两者大小一致（可能会由于四舍五入原因不一致）
alpha = cv2.resize(alpha, (bgRoi.shape[1], bgRoi.shape[0]))
bg = cv2.multiply(alpha, bgRoi)
bg = bg.astype('uint8')
# 提取帽子区域
hat = cv2.bitwise_and(resizedHat, cv2.bitwise_not(maskInv))
# 相加之前保证两者大小一致（可能会由于四舍五入原因不一致）")
hat = cv2.resize(hat, (bgRoi.shape[1], bgRoi.shape[0]))
# 两个ROI区域相加")
addHat = cv2.add(bg, hat)
# 把添加好帽子的区域放回原图
img[y + dh - resizedHatH:y + dh,
(eyes_center[0] -
resizedHatW // 3):(eyes_center[0] +
resizedHatW // 3 * 2)] = addHat
return img
