def get_board_boundaries(self, image):
"""
Function gets rid of low intensity pixel values in
color spectrum and sets them to 0. Then it calculates
histogram of green color in x and y axis from image.
Then we calculate 1-th order difference on this
histogram
Args:
image(numpy.ndarray): Loaded image
Retunrs:
Returns extremes from calculated histogram
"""
(width, height, _) = image.shape
vlines = np.zeros(width)
hlines = np.zeros(height)
image[:, :, 1] = cv.inRange(image[:, :, 1], self.lo_thresh,
self.hi_thresh)
vlines = cv.reduce(image[:, :, 1],
1,
cv.cv.CV_REDUCE_SUM,
dtype=cv.CV_32S)
hlines = cv.reduce(image[:, :, 1],
0,
cv.cv.CV_REDUCE_SUM,
dtype=cv.CV_32S)
vlines = map(lambda *row: list(row), *vlines)
y = np.diff(vlines[0])
x = np.diff(hlines[0])
return (self.__get_extremes(x), self.__get_extremes(y))
def _find_pipe(self):
havg = cv2.reduce(self.field_image, 0, cv2.cv.CV_REDUCE_AVG)[0]
for x in xrange(len(havg)):
if havg[x] > kPipePixelThreshold:
continue
# Find width
width = 0
while (x + width < len(havg)) and (havg[x + width] <= kPipePixelThreshold):
width += 1
if not (kPipeMinWidth <= width <= kPipeMaxWidth):
x += width
continue
# Find opening
pipe_slice = self.field_image[:, x:x + width]
vavg = cv2.reduce(pipe_slice, 1, cv2.cv.CV_REDUCE_AVG)
def find_first_gap_row(vavg):
for y in xrange(len(vavg)):
if vavg[y][0] > kPipePixelThreshold:
return y
return None
top_h = find_first_gap_row(vavg)
bottom_h = find_first_gap_row(vavg[::-1])
if top_h is None or bottom_h is None:
continue
if len(vavg) - top_h - bottom_h < kPipeMinOpening:
continue
return x - 2, width + 4, top_h, bottom_h
return None, None, None, None
def trim_image(image, padding=10):
nr_rows = image.shape[0]
nr_cols = image.shape[1]
rows_reduced = cv2.reduce(image, 1, cv2.REDUCE_MIN).reshape(-1)
cols_reduced = cv2.reduce(image, 0, cv2.REDUCE_MIN).reshape(-1)
# First, determine the number of blanc lines from the top
threshold = 250
blancs_top = count_blancs_until_component(rows_reduced,
min_threshold=threshold)
new_top = max(blancs_top - padding, 0)
blancs_bottom = count_blancs_until_component(rows_reduced[::-1],
min_threshold=threshold)
new_bottom = min(nr_rows - blancs_bottom + padding, nr_rows)
blancs_left = count_blancs_until_component(cols_reduced,
min_threshold=threshold)
new_left = max(blancs_left - padding, 0)
blancs_right = count_blancs_until_component(cols_reduced[::-1],
min_threshold=threshold)
new_right = min(nr_cols - blancs_right + padding, nr_cols)
# Crop image
trimmed_image = image[new_top:new_bottom, new_left:new_right]
return trimmed_image
def get_board_boundaries(self, image):
"""
Function gets rid of low intensity pixel values in
color spectrum and sets them to 0. Then it calculates
histogram of green color in x and y axis from image.
Then we calculate 1-th order difference on this
histogram
Args:
image(numpy.ndarray): Loaded image
Retunrs:
Returns extremes from calculated histogram
"""
(width, height, _) = image.shape
vlines = np.zeros(width)
hlines = np.zeros(height)
image[:, :, 1] = cv.inRange(image[:, :, 1],
self.lo_thresh,
self.hi_thresh)
vlines = cv.reduce(image[:, :, 1], 1,
cv.cv.CV_REDUCE_SUM,
dtype=cv.CV_32S)
hlines = cv.reduce(image[:, :, 1], 0,
cv.cv.CV_REDUCE_SUM,
dtype=cv.CV_32S)
vlines = map(lambda *row: list(row), *vlines)
y = np.diff(vlines[0])
x = np.diff(hlines[0])
return (self.__get_extremes(x), self.__get_extremes(y))
def words_seperator(lower, upper, cnt_lines):
hist2 = cv2.reduce(rotated[upper:lower, :], 0, cv2.REDUCE_AVG).reshape(-1)
left = 0
right = len(hist2) - 1
while hist2[left] <= min(hist2):
left += 1
while hist2[right] <= min(hist2):
right -= 1
x = left
prev_x = left
cnt_words = 1
while x < right:
while x < len(hist2) and hist2[x] <= min(hist2):
x += 1
prev_x = x
while x < len(hist2) and hist2[x] > min(hist2):
x += 1
if x - prev_x > 5:
cv2.line(output, (prev_x, lower), (prev_x, upper), (255, 0, 0))
cv2.line(output, (x, lower), (x, upper), (255, 0, 0))
hist3 = cv2.reduce(
rotated[upper - BUFFER:lower + BUFFER,
prev_x - BUFFER:x + BUFFER], 1,
cv2.REDUCE_AVG).reshape(-1)
letter_upper = 0
letter_lower = len(hist3) - 1
while hist3[letter_upper] <= min(hist3):
letter_upper += 1
while hist3[letter_lower] <= min(hist3):
letter_lower -= 1
cv2.imwrite(
"word/words_" + str(cnt_lines) + str(cnt_words) + ".png",
output2[upper + letter_upper - BUFFER:upper + letter_lower +
BUFFER, prev_x - BUFFER:x + BUFFER])
cnt_words += 1
def findpeaks(image, dist, horizontal_profile=True):
if (horizontal_profile):
intensity_hist = cv2.reduce(image, 1, cv2.REDUCE_SUM, dtype=cv2.CV_32S)
else:
intensity_hist = cv2.reduce(image, 0, cv2.REDUCE_SUM, dtype=cv2.CV_32S).T
## smoothen the intensity histogram using savgol to avoid noise
x = np.reshape(intensity_hist, len(intensity_hist))
window_size = 5
if (horizontal_profile):
window_size = int(image.shape[0]/22) ## magic num!
else:
window_size = int(image.shape[1]/22) ## magic num!
if (window_size%2 == 0):
window_size = window_size+1
intensity_hist_s = ss.savgol_filter(x, window_size, 5)
intensity_hist_s = intensity_hist_s.astype(int)
## find peaks in smooth plot. Prominence is via observation - if a line has just
## a few letters it will create a small peak. We enforce roughly 1/10 of the line
## should be occupied with letters
## distance is to avoid quirks. We use 5 degree poly to smoothen so it leads to dual
## peaks at times close together. Distance specification is a heuristic to avoid
## getting too close peaks. Ideally we should skip that, but use this routine recursively
## by plugging in the median of linewidths back into the find_peaks as distance till
## convergence
maxval = np.max(intensity_hist_s)
peaks, _ = ss.find_peaks(intensity_hist_s, prominence=maxval/10, distance=dist)
return intensity_hist, peaks, intensity_hist_s
def getEqnLayout(self, thresh, img):
H,W = thresh.shape[:2]
hist = cv2.reduce(thresh, 0, cv2.REDUCE_AVG).reshape(-1)
th = 0
lefts = [y for y in range(W-1) if hist[y]<=th and hist[y+1]>th]
rights = [y for y in range(W-1) if hist[y]>th and hist[y+1]<=th]
uppers = []
lowers = []
for i in range(0,len(lefts)):
temp = np.copy(thresh[:, lefts[i]:rights[i] + 1])
H,W = temp.shape[:2]
hist = cv2.reduce(temp, 1, cv2.REDUCE_AVG).reshape(-1)
th = 0
uppers.append([y for y in range(H-1) if hist[y]<=th and hist[y+1]>th])
lowers.append([y for y in range(H-1) if hist[y]>th and hist[y+1]<=th])
chars = []
numOfClumps = {}
for i in range(0, len(lefts)):
for j in range(0, len(uppers[i])):
numOfClumps[(i,j)] = 0
temp = np.copy(thresh[uppers[i][j]:lowers[i][j]+1, lefts[i]:rights[i]+1])
contours= cv2.findContours(temp, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[1]
rects = [cv2.boundingRect(c) for c in contours]
rects = sorted(rects,key=lambda l:l[0])
for k in range(0, len(rects)):
rects[k] = rects[k][0], rects[k][1], rects[k][0]+rects[k][2],\
rects[k][1]+rects[k][3]
k = 0
while k < len(rects)-1:
xmin1, xmax1 = rects[k][1], rects[k][3]
xmin2, xmax2 = rects[k+1][1], rects[k+1][3]
ymin1, ymax1 = rects[k][0], rects[k][2]
ymin2, ymax2 = rects[k+1][0], rects[k+1][2]
#print(xmin1, xmax1, xmin2, xmax2, ymin1, ymax1, ymin2, ymax2)
if (ymin1 <= ymin2 and ymax2 <= ymax1 or ymin2 <= ymin1 and ymax1 <= ymax2) and \
(abs(max(xmin1,xmin2) - min(xmax1,xmax2)) / \
abs(max(xmax1, xmax2) - min(xmin1,xmin2)) < 0.50) and \
((xmax1-xmin1 <= 25 and ymax1-ymin1 <= 25) or \
(xmax2-xmin2 <= 25 and ymax2-ymin2 <= 25)):####:
rects[k] = list((min(ymin1, ymin2), min(xmin1, xmin2),
max(ymax1,ymax2), max(xmax1, xmax2)))
del rects[k+1]
k += 1
for cnt in range(0, len(rects)):
numOfClumps[(i,j)] += 1
chars.append(list(((i,j,cnt), lefts[i]+rects[cnt][0], uppers[i][j]+rects[cnt][1],\
lefts[i]+rects[cnt][2], uppers[i][j]+rects[cnt][3]))) #doubt
'''cv2.rectangle(img, (chars[-1][1], chars[-1][2]), (chars[-1][3], chars[-1][4]), \
(0, 0, 255), 1)
cv2.imwrite('ltx/z0.png', img) '''
return lefts, rights, uppers, lowers, numOfClumps, chars
def crop(img):
'''
co=np.column_stack(np.where(img>0))
x,y,w,h=cv2.boundingRect(co)
img=img[x:x+w,y:y+h]
'''
y_sum = cv2.reduce(img, 0, cv2.REDUCE_SUM, dtype=cv2.CV_32S)[0]
i = 0
j = img.shape[1] - 1
y_norm = [(x) / max(y_sum) for x in y_sum]
while (y_norm[i] <= 0.1 and i < j):
i = i + 1
while (y_norm[j] <= 0.1 and j > 0):
j = j - 1
x_sum = cv2.reduce(img, 1, cv2.REDUCE_SUM, dtype=cv2.CV_32S)
x_norm = [(x) / max(x_sum) for x in x_sum]
k = 0
l = img.shape[0] - 1
while (x_norm[k][0] <= 0.1 and k < l):
k = k + 1
while (x_norm[l][0] <= 0.2 and l > 0):
l = l - 1
img1 = img[k:l, i:j]
#cv2.imshow("a",resize(img1,500,500))
#cv2.waitKey(0)
return img1
def image_projection(img):
"""
"""
# change to gray
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# Change to numpy array format
nb = np.array(gray)
x_sum = cv2.reduce(gray, 0, cv2.REDUCE_SUM, dtype=cv2.CV_64F)
y_sum = cv2.reduce(gray, 1, cv2.REDUCE_SUM, dtype=cv2.CV_64F)
# get height and weight
x = gray.shape[1]
y = gray.shape[0]
# division the result by height and weight
x_sum = x_sum / y
y_sum = y_sum / x
x_sum = x_sum.transpose()
x_arr = np.arange(x)
y_arr = np.arange(y)
xn = np.array(x_sum)
yn = np.array(y_sum)
x_peaks, _ = find_peaks(xn.ravel(), height=0, width=10)
y_peaks, _ = find_peaks(yn.ravel(), height=0, width=10)
return x_arr, xn, x_peaks, y_arr, yn, y_peaks
def histogram(self):
"""Add histograms to an image.
A new stage of the image, *histogram*, is added.
The histograms contain an indication of the amount of ink in
horizontal and in vertical rows.
The histogram values are preserved as well.
"""
if self.empty:
return
batch = self.batch
boxed = self.boxed
stages = self.stages
rotated = self.stages["rotated"]
self.histX = cv2.reduce(rotated, 0, cv2.REDUCE_AVG).reshape(-1)
self.histY = cv2.reduce(rotated, 1, cv2.REDUCE_AVG).reshape(-1)
if not batch or boxed:
normalizedC = stages["normalizedC"]
if not batch:
histogram = normalizedC.copy()
for (hist, vert) in ((self.histY, True), (self.histX, False)):
for (i, val) in enumerate(self.histY):
color = (int(val), int(2 * val), int(val))
index = (0, i) if vert else (i, 0)
value = (val, i) if vert else (i, val)
cv2.line(histogram, index, value, color, 1)
stages["histogram"] = histogram
def get_difference(image1, image2):
'''
Get difference between two images
'''
diff = image1.copy()
cv.absdiff(image1, image2, diff)
reduction = cv.reduce(diff, 0, cv.REDUCE_AVG)
reduction = cv.reduce(reduction, 1, cv.REDUCE_AVG)
return reduction
def cropMoney(inputImage):
"""
main function for RoI pre-processing
:param inputImage: RMB bill image
:return: resultImage: 2 PotentialRoIs from one side RMB bill
"""
#find horizontal cutting point
inputImage2 = np.copy(
inputImage) #copy to second var to keep the originality
inputImage2[inputImage2 < 80] = 0
verProjectedMoney = cv2.reduce(inputImage2,
0,
cv2.REDUCE_SUM,
dtype=cv2.CV_32S)
normalized = np.copy(verProjectedMoney)
cv2.normalize(verProjectedMoney, normalized, 0, 255, cv2.NORM_MINMAX)
normalized = normalized.astype(np.uint8)
diff = abs(np.diff(normalized.astype(np.int32), 1, axis=1)).flatten()
loc = np.argwhere(diff > 4)
minHor = np.asscalar(np.min(loc))
maxHor = np.asscalar(np.max(loc))
#find vertical cutting point
horProjectedMoney = cv2.reduce(inputImage2,
1,
cv2.REDUCE_SUM,
dtype=cv2.CV_32S)
normalized = np.copy(horProjectedMoney)
cv2.normalize(horProjectedMoney, normalized, 0, 255, cv2.NORM_MINMAX)
normalized = normalized.astype(np.uint8)
diff = abs(np.diff(normalized.astype(np.int32), 0, axis=1)).flatten()
loc = np.argwhere(diff > 50)
minVer = np.asscalar(np.min(loc))
maxVer = np.asscalar(np.max(loc))
croppedImage = inputImage[minVer:maxVer, minHor:maxHor]
widthScalledMoney = 1068
heightScalledMoney = 420
croppedImage = cv2.resize(croppedImage,
(widthScalledMoney, heightScalledMoney))
flipped = np.flip(np.flip(np.copy(croppedImage), axis=0), axis=1)
flipped = flipped.astype(np.uint8)
minHorPotROI = potRoiCoordinate[0]
minVerPotROI = potRoiCoordinate[1]
maxHorPotROI = potRoiCoordinate[2]
maxVerPotROI = potRoiCoordinate[3]
potROI = np.copy(croppedImage[minVerPotROI:maxVerPotROI,
minHorPotROI:maxHorPotROI])
potRoiFlipped = np.copy(flipped[minVerPotROI:maxVerPotROI,
minHorPotROI:maxHorPotROI])
potROI = np.expand_dims(potROI, axis=0)
potRoiFlipped = np.expand_dims(potRoiFlipped, axis=0)
PotentialRoIs = np.concatenate((potROI, potRoiFlipped), axis=0)
croppedImage = np.expand_dims(croppedImage, axis=0).astype(np.uint8)
return PotentialRoIs, croppedImage
def start_detection(images):
result.config(text="NO BREAKAGE")
img = cv2.imread(images)
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
edges = cv2.Canny(gray,50,150,apertureSize = 3)
edges = cv2.GaussianBlur(edges,(5,5),0)
k_square = np.ones((5,5),np.uint8)
k_ellipse = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(5,5))
thresh = cv2.dilate(edges,k_ellipse,iterations=1)
thresh = cv2.erode(thresh,k_square,iterations=2)
minLineLength = 100
maxLineGap = 10
lines = cv2.HoughLinesP(thresh,1,np.pi/180,100,minLineLength,maxLineGap)
x_sum = cv2.reduce(thresh, 0, cv2.REDUCE_SUM, dtype=cv2.CV_32S)
y_sum = cv2.reduce(thresh, 1, cv2.REDUCE_SUM, dtype=cv2.CV_32S)
if lines is not None:
for line in lines:
for x1,y1,x2,y2 in line:
cv2.line(img,(x1,y1),(x2,y2),(0,255,0),2)
h=img.shape[0]
w= img.shape[1]
flag1=0
flag2=0
flag3=0
count = 0
n=0
p=0
for i in range(0,h-5,5):
flag1=0
flag2=0
flag3=0
for j in range(0,w-5,5):
count=0;
for k in range(i,i+5):
for l in range(j,j+5):
p=n
if img[k][l][0]==0 and img[k][l][1]==255 and img[k][l][2]==0:
n=j
#count=count+1
#print(i," ",j," ",img[k][l]," ")
if (n-p)>30:
flag3=1
break
if flag3==1:
break
if flag3==1:
result.config(text="BREAKAGE DETECTED")
ima=images.replace('.jpg','res.jpg')
cv2.imwrite(ima,img)
pro_image(ima)
def segment(image, output_directory, padding=10, runmode=1):
lines_directory = output_directory + '/lines'
words_li_li = []
# 1 = column reduction.
# CV_REDUCE_AVG instead of sum, because we want the normalized number of pixels
histogram = cv2.reduce(image, 1, cv2.REDUCE_AVG)
# Transpose column vector into row vector
histogram = histogram.reshape(-1)
lookahead = 30
line_peaks = peakdetect(histogram, lookahead=lookahead)
remove_directory(lines_directory)
ensure_directory(lines_directory)
lines = segment_image_into_lines(image,
line_peaks,
lines_directory,
padding=padding,
runmode=runmode)
line_idx = 0
for line in lines:
line_histogram = cv2.reduce(line, 0, cv2.REDUCE_AVG)
# Transpose column vector into row vector
line_histogram = line_histogram.reshape(-1)
lookahead = 30
word_peaks = peakdetect(line_histogram, lookahead=lookahead)
if runmode > 1:
plt.plot(line_histogram)
plt.title('Line=' + str(line_idx))
plt.savefig(lines_directory + '/histogram_' + str(line_idx) +
'.jpg')
plt.clf()
words_directory = lines_directory + '/line_' + str(line_idx)
remove_directory(words_directory)
ensure_directory(words_directory)
words = segment_line_into_words(line,
line_idx,
word_peaks,
words_directory,
padding=padding,
runmode=runmode)
line_idx = line_idx + 1
words_li_li.append(words)
return words_li_li
def find_notes(staff_crops, valid_blobs):
note_rects = [[] for _ in range(valid_blobs)]
for i in range(valid_blobs):
staff_crop = staff_crops[valid_blobs - 1 - i]
x_distribution = cv2.reduce(staff_crop,
0,
cv2.REDUCE_SUM,
dtype=cv2.CV_32S)[0] # column sum
min = staff_crop.shape[1] * 255
x_1 = -1
for val in x_distribution[50:len(x_distribution) - 50]:
if val < min:
min = val
min += 1800
# print 'notes %d' % (i+1)
# print "min: %d" % min
for j in range(len(x_distribution)):
val = x_distribution[j]
if val < min and x_1 != -1 and j - x_1 > 5:
# print "ENDS from %d to %d: %d" % (x_1, j, val)
longest_consecutive = 0
count = 0
y_start = 0
y_end = staff_crop.shape[0]
note_horizontal_crop = staff_crop[0:staff_crop.shape[0], x_1:j]
y_distribution = np.array(
cv2.reduce(note_horizontal_crop,
1,
cv2.REDUCE_SUM,
dtype=cv2.CV_32S).flatten()) # row sum
for k in range(len(y_distribution)):
row_sum = y_distribution[k]
if row_sum > 0:
if count == 0:
y_start_temp = k
count += 1
else:
if count > longest_consecutive:
longest_consecutive = count
y_start = y_start_temp
y_end = k
count = 0
note_rects[valid_blobs - 1 - i].append(
((x_1, y_start), (j, y_end)))
x_1 = -1
elif val > min and x_1 == -1:
# print "STARTS at %d: %d" % (j, val)
x_1 = j
return note_rects
def fixRotation(image):
mask = np.uint8(np.where(image >= 10, 0, 1))
row_counts = cv2.reduce(mask, 1, cv2.REDUCE_SUM, dtype=cv2.CV_32SC1)
col_counts = cv2.reduce(mask, 0, cv2.REDUCE_SUM, dtype=cv2.CV_32SC1)
rows = row_counts.flatten().tolist()
cols = col_counts.flatten().tolist()
up = sum(rows[:5])
down = sum(rows[-5:])
left = sum(cols[:5])
right = sum(cols[-5:])
if (left > up and left > down) or (right > up and right > down):
image = ndimage.rotate(image, 90)
return image
def resize_to_1080(screen,
crop_bars: bool,
interpolation=cv2.INTER_NEAREST,
respect_width=True):
if crop_bars:
gimg = cv2.cvtColor(screen, cv2.COLOR_BGR2GRAY)
hline = np.ravel(cv2.reduce(gimg, 0, cv2.REDUCE_SUM,
dtype=cv2.CV_32S)) > 0
vline = np.ravel(
cv2.reduce(gimg[:, 500:], 1, cv2.REDUCE_SUM, dtype=cv2.CV_32S)) > 0
topbar = np.argmax(vline)
if topbar > 5:
bottombar = np.argmax(vline[-5::-1]) + 4
else:
bottombar = np.argmax(vline[::-1])
leftbar, rightbar = np.argmax(hline), np.argmax(hline[::-1])
if screen.shape[0] - (topbar + bottombar) > 250 and screen.shape[1] - (
leftbar + rightbar) > 250:
screen = screen[topbar:screen.shape[0] - bottombar,
leftbar:screen.shape[1] - rightbar, ]
if respect_width:
if screen.shape[0] == 1080:
return screen
scale = 1080 / screen.shape[0]
if screen.shape[1] * scale > 2560:
scale = 2560 / screen.shape[1]
cscale = min(scale, 2)
scaled = cv2.resize(screen, (0, 0),
fx=cscale,
fy=cscale,
interpolation=interpolation)
if scale <= 2:
return scaled
else:
return cv2.copyMakeBorder(
scaled,
0,
max(1080 - scaled.shape[0], 0),
0,
max(1920 - scaled.shape[1], 0),
cv2.BORDER_CONSTANT,
value=(0, 0, 0),
)
else:
return cv2.resize(screen, (1920, 1080), interpolation=interpolation)
def calculateDifference(imageA, imageB, width, height):
d = 0
vectorA = cv2.reduce(imageA, 0, cv2.REDUCE_SUM, dtype=cv2.CV_32S)
vectorB = cv2.reduce(imageB, 0, cv2.REDUCE_SUM, dtype=cv2.CV_32S)
for i in range(len(vectorA[0])):
pixelA = vectorA[0][i]
pixelB = vectorB[0][i]
db = pixelA[0] - pixelB[0]
dg = pixelA[1] - pixelB[1]
dr = pixelA[2] - pixelB[2]
d = d + gmpy2.isqrt(dr * dr + dg * dg + db * db)
return d
def get_histogram_for_angle(self, image, angle):
"""
Rotates an image for the specified angle and calculates
sum of pixel values for every row
Args:
image: PIL image object
angle: rotation angle in degrees
Returns:
list of values. Each value is a sum of inverted pixel's
for the corresponding row
"""
copy = image.rotate(angle, Image.BILINEAR, True)
img = np.fromstring(copy.tostring(),
dtype=np.uint8).reshape(copy.size[1],
copy.size[0], 2)
alpha = img[:, :, 1]
res = img[:, :, 0]
res[res >= MAGIC_COLOR_THRESHOLD] = 255
res[alpha < 255] = 255
res = 255 - res
# Python cv2.reduce doesn't work correctly with int matrices.
data_for_reduce = res.astype(np.float)
histogram = cv2.reduce(data_for_reduce, 1, cv2.cv.CV_REDUCE_SUM)[:, 0]
return histogram
def LinesMedian(bw_image): # making horizontal projections horProj = cv2.reduce(bw_image, 1, cv2.cv.CV_REDUCE_AVG) # make hist - same dimension as horProj - if 0 (space), then True, else False th = 0; # black pixels threshold value. this represents the space lines hist = horProj <= th; #Get mean coordinate of white white pixels groups ycoords = [] y = 0 count = 0 isSpace = False median_count = [] for i in range(0, bw_image.shape[0]): if (not isSpace): if (hist[i]): #if space is detected, get the first starting y-coordinates and start count at 1 isSpace = True count = 1 #y = i else: if (not hist[i]): isSpace = False median_count.append(count) else: #y = y + i count = count + 1 median_count.append(count) #ycoords.append(y / count) #returns counts of each blank rows of each of the lines found return median_count
def histogramaHor(self):
self.plotHisto.clear()
#imag3=self.img
# Conversion a escala de grices
imag3 = cv2.cvtColor(self.img, cv2.COLOR_BGR2GRAY)
# Calculo de Histograma Horizontal
x_sum = cv2.reduce(imag3, 0, cv2.REDUCE_SUM, dtype=cv2.CV_32S)
x_sum_len = len(x_sum[0])
x_sum_norm = x_sum[0] / (x_sum_len)
x_sum_norm = x_sum_norm.round(0)
x_sum_norm = x_sum_norm.astype(int)
print(x_sum_norm)
x_sum_x = np.linspace(0, 255, 256)
x_sum_x = x_sum_x.astype(int)
x_sum_y = np.zeros(256)
for i in range(256):
for j in range(x_sum_len):
if (x_sum_norm[j] == i):
x_sum_y[i] += 1
#pg.plot(x_sum_x, x_sum_y, title=('Histograma Horizontal'))
self.plotHisto.plot(x_sum_x, x_sum_y, title=('Histograma Horizontal'))
self.plotHisto.setLabel('left', 'Cantidad de pixeles')
self.plotHisto.setLabel('bottom', 'Intensidad de iluminacion')
def findLines(bw_image, LinesThres):
horProj = cv2.reduce(bw_image, 1,
cv2.REDUCE_AVG) # making horizontal projections
# make hist - same dimension as horProj - if 0 (space), then True, else False
th = 0 # black pixels threshold value. this represents the space lines
hist = horProj <= th
#Get mean coordinate of white white pixels groups
ycoords = []
y = 0
count = 0
isSpace = False
for i in range(0, bw_image.shape[0]):
if (not isSpace):
if (
hist[i]
): #if space is detected, get the first starting y-coordinates and start count at 1
isSpace = True
count = 1
y = i
else:
if (not hist[i]):
isSpace = False
#when smoothing, thin letters will breakdown, creating a new blank lines or pixel rows, but the count will be small, so we set a threshold.
if (count >= LinesThres):
ycoords.append(y / count)
else:
y = y + i
count = count + 1
ycoords.append(y / count)
#returns y-coordinates of the lines found
return ycoords
def LinesMedian(bw_image):
horProj = cv2.reduce(bw_image, 1,
cv2.REDUCE_AVG) # making horizontal projections
# make hist - same dimension as horProj - if 0 (space), then True, else False
th = 0 # black pixels threshold value. this represents the space lines
hist = horProj <= th
#Get mean coordinate of white white pixels groups
count = 0
isSpace = False
median_count = []
for i in range(0, bw_image.shape[0]):
if (not isSpace):
if (
hist[i]
): #if space is detected, get the first starting y-coordinates and start count at 1
isSpace = True
count = 1
else:
if (not hist[i]):
isSpace = False
median_count.append(count)
else:
count = count + 1
median_count.append(count)
#returns counts of each blank rows of each of the lines found
return median_count
def segment_lines(im):
im_proj = cv2.reduce(im, 1, cv2.REDUCE_AVG)
hist = im_proj == 255
ycoords = []
y = 0
count = 0
isSpace = False
for i in range(len(hist)):
if not isSpace:
if hist[i]:
isSpace = True
count = 1
y = i
else:
if not hist[i]:
isSpace = False
ycoords.append(y / count)
else:
y += i
count += 1
lines = [
line for line in np.vsplit(im,
np.array(ycoords).astype(int))
if not (line == 255).all()
]
return lines
def processLine(i):
cv2.namedWindow("result", cv2.WINDOW_NORMAL)
subimg = img_bw[uppers[i]:lowers[i+1], 0:W]
subimg_c = aligned[uppers[i]:lowers[i+1], 0:W]
# break into characters via histogramming trick.
hist = cv2.reduce(subimg, 0, cv2.REDUCE_AVG).reshape(-1)
hist = cv2.GaussianBlur(hist,(5,5),0)
#1300 is magic empirical constant
char = {'l':[], 'r':[]}
def bindUpdateBounds(ratio):
(temp1, temp2) = updateBounds(ratio, lhist=hist, mimg = subimg_c, dim=0)
print(temp1)
char['l'] = temp1;
char['r'] = temp2;
cv2.createTrackbar("ratio", "result", 1300, 2000, bindUpdateBounds)
bindUpdateBounds(1300)
cv2.waitKey()
cv2.destroyAllWindows()
# armed with individual character images, standardize them if possible
# then export to folders.
for j, _ in enumerate(char['l']):
print("Running character processing")
processChar(char['r'][j], char['l'][j], subimg, lowers[i+1]-uppers[i])
line= ""
print(line)
return line
def separate_words(line):
line_hist = cv2.reduce(line, 0, cv2.REDUCE_AVG).reshape(-1)
new_line = cv2.cvtColor(line, cv2.COLOR_GRAY2BGR)
line_peaks = peakdetect(line_hist, lookahead=50)
Hl, Wl = new_line.shape[:2]
words = []
for y in line_peaks[0]:
if y[1] == 255:
words.append(y[0])
# plt.plot(y[0], y[1], "r*")
if y[1] == 255:
cv2.line(new_line, (y[0], 0), (y[0], Hl), (255, 0, 0), 3)
# for y in line_peaks[1]:
# plt.plot(y[0], y[1], "g*")
# if y[1] == 255:
# words.append(y[0])
# cv2.line(new_line, (y[0], 0), (y[0], Hl), (0, 255, 0), 3)
words.insert(0, 0)
words.append(Wl)
# plt.imshow(new_line, cmap=plt.cm.gray)
# plt.show()
return words
def deskew(im):
d_angle = 10
c_angle = 0
for delta in [10, 1, 0.1]:
angles = np.arange(c_angle - d_angle, c_angle + d_angle, delta)
scores = []
for angle in angles:
data = interpolation.rotate(im,
angle,
reshape=False,
order=0,
cval=255)
#hist = np.sum(data, axis=1)
hist = cv2.reduce(data, 1, cv2.REDUCE_AVG)
score = np.std(hist)**2
scores.append(score)
best_score = max(scores)
c_angle = best_angle = angles[scores.index(best_score)]
d_angle = delta
im_rot = interpolation.rotate(im,
best_angle,
reshape=False,
order=0,
cval=255)
return im_rot
def is_Stripe(image, number_peaks=12): # -> Boolean
""" Checks if the image is striped.
@Parameters: image matrix (binary form - black and white)
number_peaks (int) Number of peaks to be considered 'striped'
@Returns Boolean, if striped.
"""
# Crop the edges
image = image[4:-4]
# Use the Sobel operator across the Y to obtain the Gradient
fpx = cv2.Sobel(image, cv2.CV_64F, 0, 1, ksize=5) # vertical
# Project it onto single dimension (1D) - by summing them
vpr = cv2.reduce(fpx, 1, cv2.REDUCE_SUM)
# Squaring and square rooting removes the negative sign, and flatten
vpr_p = np.sqrt(vpr**2).flatten()
# Large spikes on f'(x) indicate change in the rate of change - seek areas where slope of the second derivative is 0
unfiltered_peaks, _ = find_peaks(vpr_p, height=0) # Finding peaks
# Filter the peaks for noise which will be defined as any peak with an absolute height of less than PEAK_THRESH * maximum peak length
all_peaks = (vpr_p[i] for i in unfiltered_peaks)
peak_thresh = max(all_peaks) * PEAK_THRESH
filtered_peaks = sum(1 for i in unfiltered_peaks
if vpr_p[i] >= peak_thresh)
return (filtered_peaks) > number_peaks
def separate_cha(line):
line_hist = cv2.reduce(line, 0, cv2.REDUCE_AVG).reshape(-1)
new_line = cv2.cvtColor(line, cv2.COLOR_GRAY2BGR)
line_peaks = peakdetect(line_hist, lookahead=25)
Hl, Wl = new_line.shape[:2]
cha = []
# for y in line_peaks[0]:
# plt.plot(y[0], y[1], "r*")
# cv2.line(new_line, (y[0], 0), (y[0], Hl), (255, 0, 0), 3)
for y in line_peaks[0]:
if y[1] >= 235:
cha.append(y[0])
# plt.plot(y[0], y[1], "g*")
cv2.line(new_line, (y[0], 0), (y[0], Hl), (0, 255, 0), 3)
cha.insert(0, 0)
cha.append(Wl)
# plt.plot(line_hist)
# plt.show()
# plt.imshow(new_line, cmap=plt.cm.gray)
# plt.show()
return cha
def prepareImage(img):
w, h = 19, 19
img = img[:h, :w]
hist = cv2.reduce(img, 0, cv2.REDUCE_AVG).reshape(-1)
maxpoint = np.argmin(hist)
newimg = np.roll(img, (0, w // 2 - maxpoint))
return np.array(newimg > 150, dtype='int').flatten()
def ConvertToBinaryImage(self, imagePath):
"""
Open the original image, convert it to binary, save the binary image in the sam directory
of the originak one and return the path to the binary image.
:param imagePath: path to the image we want to convert into binary image
:type imagePath: string
"""
img = cv.imread(imagePath, 0)
ret, binaryImage = cv.threshold(img, 127, 255, cv.THRESH_BINARY_INV)
## (3) minAreaRect on the nozeros
pts = cv.findNonZero(binaryImage)
ret = cv.minAreaRect(pts)
(cx, cy), (w, h), ang = ret
if w > h:
w, h = h, w
ang += 90
M = cv.getRotationMatrix2D((cx, cy), 0, 1.0)
rotated = cv.warpAffine(binaryImage, M, (img.shape[1], img.shape[0]))
hist = cv.reduce(rotated, 1, cv.REDUCE_AVG).reshape(-1)
th = 2
H, W = img.shape[:2]
uppers = [y for y in range(H - 1) if hist[y] <= th and hist[y + 1] > th]
lowers = [y for y in range(H - 1) if hist[y] > th and hist[y + 1] <= th]
rotated = cv.cvtColor(rotated, cv.COLOR_GRAY2BGR)
for y in uppers:
cv.line(rotated, (0, y), (W, y), (255, 0, 0), 1)
for y in lowers:
cv.line(rotated, (0, y), (W, y), (0, 255, 0), 1)
cv.imwrite("result.png", rotated)
# Add "_bin" to the name of the binary image.
parts = imagePath.split(".")
binaryImagePath = parts[0] + "_bin." + parts[1]
# Save the binary image in the same directory.
cv.imwrite(binaryImagePath, binaryImage)
image = cv.imread(binaryImagePath)
uppers, lowers = self.boundingHandle(uppers, lowers)
i = 0
directory = None
while i != len(uppers):
# create help folder named "row" and write each row
file_path = "C:/rows/" + str(i) + '.png'
directory = os.path.dirname(file_path)
try:
os.stat(directory)
except:
os.mkdir(directory)
y = abs(uppers[i] - 20)
x = 0
h = (lowers[i] - uppers[i]) + 40
row = image[y:y + h, x:x + W]
cv.imwrite(file_path, row)
i += 1
os.remove(binaryImagePath)
# Return the path to the directory of binary images.
return directory
def train(self):
ustep = self.range_dist[0]/self.hist_bins[0]
vstep = self.range_dist[1]/self.hist_bins[1]
#calc n, X_i and mu
mu=np.array((1,2))
n = cv2.countNonZero(self.f_hist);
count = 0;
N = 0;
X=np.array((n,2))
f=[]
for ubin in self.hist_bins[0]:
for vbin in self.hist_bins[1]:
histval = self.f_hist[ubin][vbin];
if histval > 0:
sampleX = ((1,2) << self.low_range[0] + self.ustep * (ubin+.5), self.low_range[1] + vstep * (vbin+.5))
count=count+1
sampleX=X.row[count]
f.append(histval)
mu = mu+ histval * sampleX;
N += histval
mu /= N;
#calc psi - mean of DB
self.psi=cv2.reduce(X, self.psi,0, cv.CV_REDUCE_AVG)
#calc Lambda
self.Lambda=np.zeros((2,2),np.uint)
for i in n:
X_m_mu = (X.row[i] - mu)
prod = f[i] * X_m_mu.t() * X_m_mu
self.Lambda += prod;
self.Lambda /= N;
linv = self.Lambda.inv();
self.Lambda_inv.val[0] = linv[0][0]
self.Lambda_inv.val[1] = linv[0][1]
self.Lambda_inv.val[2] = linv[1][0]
def findline(img):
'''
Parameters
----------
img : Array of uint8
DESCRIPTION.
Returns
-------
uppers : LIST
list of upper boundary of text.
lowers : LIST
list of lower boundary of text.
'''
# find and draw the upper and lower boundary of each lines
hist = cv2.reduce(img, 1, cv2.REDUCE_AVG).reshape(-1)
th = 2
uppers = [y for y in range(H - 1) if hist[y] <= th and hist[y + 1] > th]
lowers = [y for y in range(H - 1) if hist[y] > th and hist[y + 1] <= th]
for y in range(len(uppers)):
a = uppers[y]
b = lowers[y]
img = cv2.line(img, (0, a), (W, a), (255, 0, 0), 2)
img = cv2.line(img, (0, b), (W, b), (255, 0, 0), 2)
cv2.imshow("draw_line", img)
cv2.waitKey()
cv2.destroyAllWindows()
cv2.imwrite('draw_line.png', img)
return uppers, lowers
def LinesMedian(bw_image):
# making horizontal projections
horProj = cv2.reduce(bw_image, 1, cv2.REDUCE_AVG)
# make hist - same dimension as horProj - if 0 (space), then True, else False
th = 0
# black pixels threshold value. this represents the space lines
hist = horProj <= th
# Get mean coordinate of white white pixels groups
ycoords = []
y = 0
count = 0
isSpace = False
median_count = []
for i in range(0, bw_image.shape[0]):
if not isSpace:
if hist[i]:
isSpace = True
count = 1
# y = 1
else:
if not hist[i]:
isSpace = False
median_count.append(count)
else:
# y = y + i
count = count + 1
median_count.append(count)
# ycoords.append(y / count)
# returns counts of each blank rows of each of the lines found
return median_count
def findLines(bw_image, LinesThres):
# making horizontal projections
horProj = cv2.reduce(bw_image, 1, cv2.REDUCE_AVG)
# make hist - same dimension as horProj - if 0 (space), then True, else False
th = 0
# black pixels threshold value. this represents the space lines
hist = horProj <= th
# Get mean coordinate of white white pixels groups
ycoords = []
y = 0
count = 0
isSpace = False
for i in range(0, bw_image.shape[0]):
if not isSpace:
if hist[i]:
isSpace = True
count = 1
y = i
else:
if not hist[i]:
isSpace = False
if count >= LinesThres:
ycoords.append(y / count)
else:
y = y + i
count = count + 1
ycoords.append(y / count)
# returns y-coordinates of the lines found
return ycoords
def _find_jaw_separation_line(self, image):
'''
Finds y coordinate of lines that separates upper and lower jaws.
:param image:
:return: tuple (upper jaw line index, lower jaw line index)
'''
y_histogram = cv2.reduce(image, 1, cv2.cv.CV_REDUCE_SUM, dtype=cv2.CV_32S)
return self._get_valley_range(y_histogram)
def _get_affine_transformed_image_size(self, img_size, homo_mat):
w = img_size.width;
h = img_size.height;
corners_coordindates = [0.0, w, w, 0.0,
0.0, 0.0, h, h,
1.0, 1.0, 1.0, 1.0]
homogeneous_corners = numpy.array(corners_coordindates, dtype=numpy.float32).reshape(3, 4)
transformed_corners = numpy.dot(homo_mat, homogeneous_corners)
transformed_corners[0] = transformed_corners[0] / transformed_corners[2];
transformed_corners[1] = transformed_corners[1] / transformed_corners[2];
transformed_corners[2] = numpy.ones((1, len(transformed_corners[2])), dtype=numpy.float32)
max_corner_coord = cv2.reduce(transformed_corners, 1, cv.CV_REDUCE_MAX)
return Size(int(max_corner_coord[0, 0]), int(max_corner_coord[1, 0]))
def _find_bird(self):
hsv_image = cv2.cvtColor(self.bird_image, cv2.COLOR_BGR2HSV)
threshed = cv2.inRange(hsv_image, kBeakMinColor, kBeakMaxColor)
v = [p[0] >= 10 for p in cv2.reduce(threshed, 1, cv2.cv.CV_REDUCE_AVG)]
best = None
for y in xrange(len(v)):
for height in xrange(min(kBeakMaxLength, len(v) - y)):
if not v[y + height]:
break
if height >= kBeakMinSize:
if best is None or best[1] < height:
best = (y, height)
# cv2.imshow('bird filtered', threshed)
return None if best is None else best[0] + 8 # Aim for the center
def centralizing(self, image):
'''
With this function centralizing image in vertical direction
'''
image = np.uint8(image)
image = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
if np.count_nonzero(image) < 15:
return None
horizontal_projection = cv2.reduce(image, 1, 1)
index = 0
while (index < 28 and horizontal_projection[index] == [0]):
index += 1
image = np.delete(image, 0, 0)
index = horizontal_projection.shape[0] - 1
while (index > 10 and horizontal_projection[index] == 0):
index -= 1
image = np.delete(image, image.shape[0] - 1, 0)
(h, w) = image.shape
if w > h:
size = w
empty_line = [0] * w
rate = float(self.SIZE_OF_CHARACTER) / w
for i in xrange(w - h):
if i % 2 == 0:
image = np.vstack((image, empty_line))
else:
image = np.insert(image, 0, empty_line, axis=0)
else:
size = w
rate = float(self.SIZE_OF_CHARACTER) / w
character_image = np.zeros([size, size, 3])
character_image[:, :, 0] = image
character_image[:, :, 1] = image
character_image[:, :, 2] = image
result = cv2.resize(character_image, (0, 0), fx=rate, fy=rate)
return result
def projection_analysis(im):
# compute the ink density histogram (sum each rows)
hist = cv2.reduce(im, 1, cv2.REDUCE_SUM, dtype=cv2.CV_32F)
hist = hist.ravel()
# find peaks withing the ink density histogram
max_hist = max(hist)
mean_hist = np.mean(hist)
thres_hist = mean_hist / max_hist
peaks = peakutils.indexes(hist, thres=thres_hist, min_dist=50)
# find peaks that are too high
mean_peaks = np.mean(hist[peaks])
std_peaks = np.std(hist[peaks])
thres_peaks_high = mean_peaks + 1.5*std_peaks
thres_peaks_low = mean_peaks - 3*std_peaks
peaks = peaks[np.logical_and(hist[peaks] < thres_peaks_high,
hist[peaks] > thres_peaks_low)]
return peaks
def SpacesMedian(line): # making vertical projections verProj = cv2.reduce(line, 0, cv2.cv.CV_REDUCE_AVG) # make hist - same dimension as horProj - if 0 (space), then True, else False th = 0; # black pixels threshold value. this represents the space lines hist = verProj <= th; #Get mean coordinate of white white pixels groups xcoords = [] x = 0 count = 0 isSpace = False median_count = [] for i in range(0, line.shape[1]): if (not isSpace): if (hist[0][i]): #if space is detected, get the first starting x-coordinates and start count at 1 isSpace = True count = 1 #x = i else: if (not hist[0][i]): isSpace = False #when smoothing, thin letters will breakdown, creating a new blank lines or pixel columns, but the count will be small, so we set a threshold. #print count,"\t", #append each count of rows of blank gaps found median_count.append(count) #if (count > 15): #xcoords.append(x / count) else: #x = x + i count = count + 1 median_count.append(count) xcoords.append(x / count) #returns x-coordinates of the spaces found in the line return median_count
def get_coffee_level(img):
Y_MAX = img.shape[0]
img_coffee_level = cv2.reduce(img, 1, cv2.cv.CV_REDUCE_AVG)
debug(img_coffee_level, 'coffee_pot_reduced')
levels = [
img_coffee_level[i][0][0] for i in range(img_coffee_level.shape[0])
]
diffs = [
(i, int(levels[i]) - int(levels[i+1]))
for i in range(len(levels)-1)
]
diffs = sorted(diffs, lambda x, y: y[1] - x[1])
result = Y_MAX - diffs[0][0]
print("Coffee: %d/%d" % (result, Y_MAX))
return int(result * 100 / Y_MAX), img_coffee_level
def findLines(bw_image, LinesThres): # making horizontal projections horProj = cv2.reduce(bw_image, 1, cv2.cv.CV_REDUCE_AVG) # make hist - same dimension as horProj - if 0 (space), then True, else False th = 0; # black pixels threshold value. this represents the space lines hist = horProj <= th; #Get mean coordinate of white white pixels groups ycoords = [] y = 0 count = 0 isSpace = False for i in range(0, bw_image.shape[0]): if (not isSpace): if (hist[i]): #if space is detected, get the first starting y-coordinates and start count at 1 isSpace = True count = 1 y = i else: if (not hist[i]): isSpace = False #when smoothing, thin letters will breakdown, creating a new blank lines or pixel rows, but the count will be small, so we set a threshold. if (count >=LinesThres): ycoords.append(y / count) else: y = y + i count = count + 1 ycoords.append(y / count) #returns y-coordinates of the lines found return ycoords
import cv2 as cv
import numpy as np
data=np.genfromtxt('test.csv', delimiter=',',skip_header=1)
data_cm = np.array([]);
cv.reduce(data,data_cm,1,op = cv.CV_REDUCE_AVG)
print data_cm
#gray = np.float32(gray)
#dst = cv2.cornerHarris(gray,2,3,0.04)
ret, frame = cam.read()
fgbg = cv2.BackgroundSubtractorMOG()
while True:
ret, frame = cam.read()
if ret == True:
frame = cv2.resize(frame, (tamanoMax,tamanoMax))
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
gray = cv2.bilateralFilter(gray, 11, 17, 17)
blur = cv2.blur(gray,(10,10))
fgmask = fgbg.apply(blur) # Aplicamos la mascara
columnas = cv2.reduce(fgmask,0,cv.CV_REDUCE_MAX)
filas = cv2.reduce(fgmask,1,cv.CV_REDUCE_MAX)
for x in range(len(columnas[0])):
if columnas[0][x] > 0:
movimiento = True
break
if movimiento == False:
for x in range(len(filas)):
if filas[x,0] > 0:
movimiento = True
break
if movimiento == True and comprobarFichero():
nombre = time.strftime("%y%m%d-%H:%M:%S")+".jpg"
path="/path/donde/guardar/imagenes/"+nombre
def findFeature(self, queryImage, trainImage, trainMask=None, silent=False):
# Initiate SIFT detector
sift = cv2.xfeatures2d.SIFT_create()
# find the keypoints and descriptors with SIFT
kp_q, des_q = sift.detectAndCompute(queryImage,None)
kp_t, des_t = sift.detectAndCompute(trainImage, trainMask)
if not kp_q or not kp_t:
if not silent:
mylogger.warn("No keypoints found")
return False
# img=cv2.drawKeypoints(queryImage, kp_q, outImage=np.zeros((1,1)), flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
# cv2.imshow('img',img)
# key = cv2.waitKey(0)
# if key==1048603 or key==27:
# sys.exit()
#
# img=cv2.drawKeypoints(trainImage, kp_t, outImage=np.zeros((1,1)), flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
# cv2.imshow('img',img)
# key = cv2.waitKey(0)
# if key==1048603 or key==27:
# sys.exit()
# create BFMatcher object
bf = cv2.BFMatcher()
# Match descriptors.
matches = bf.knnMatch(des_q, des_t, k=2)
# Apply ratio test
good = []
for best_matches in matches:
if len(best_matches)>1:
m,n = best_matches
else:
m=n=best_matches[0]
if m.distance < 0.75*n.distance:
good.append([m])
if len(good)<2:
if not silent:
mylogger.warn("Feature not found")
return False
# img = cv2.drawMatchesKnn(queryImage, kp_q, trainImage, kp_t, good, outImg=np.zeros((1,1)), flags=2)
# cv2.imshow('img',img)
# key = cv2.waitKey(0)
# if key==1048603 or key==27:
# sys.exit()
img_pts = np.float32([ kp_t[m[0].trainIdx].pt for m in good ])
mean = cv2.reduce(img_pts, 0, cv2.REDUCE_AVG)[0]
# check dispersion
disp = 0
for pt in img_pts:
delta = 0
delta += (pt[0]-mean[0]) * (pt[0]-mean[0])
delta += (pt[1]-mean[1]) * (pt[1]-mean[1])
if delta>1000:
# remove extremums
np.delete(img_pts, pt)
else:
disp += delta
if len(img_pts)<2:
if not silent:
mylogger.warn("G4 not found")
return False
# recalc mean
mean = cv2.reduce(img_pts, 0, cv2.REDUCE_AVG)[0]
# mylogger.info("Dispersion: %s" % disp)
if disp>5000:
if not silent:
mylogger.warn("Dispersion too big")
return False
# cv2.circle(trainImage, (int(mean[0]), int(mean[1])), 5, (0,0,255))
# cv2.imshow('img',trainImage)
# key = cv2.waitKey(0)
# if key==1048603 or key==27:
# sys.exit()
return (int(mean[0]), int(mean[1]))
def cv_project(src):
return numpy.squeeze(cv2.reduce(src,dim=1,rtype=cv2.cv.CV_REDUCE_SUM))
def compute_sharpness_value(img_gray):
img_lap = cv2.Laplacian(img_gray, cv.CV_16S, ksize=1, scale=1, delta=0)
row_lap = cv2.reduce(img_lap, 0, cv.CV_REDUCE_MAX)
point_lap = cv2.reduce(row_lap, 1, cv.CV_REDUCE_MAX)[0][0]
return int(point_lap)
def segmentation(self):
'''
With this method segmentation text. First time detect the text line, next time the characters.
'''
self.preprocessing()
self.height, self.width = self.image.shape
size = self.width / self.NUMBER_OF_PART
text_lines = []
for i in xrange(self.NUMBER_OF_PART):
part_of_image = self.image[:, i * size:(i + 1) * size]
horizontal_projection = cv2.reduce(part_of_image, 1, 1)
histogram = horizontal_projection <= 0
histogram = histogram.astype(int)
histogram *= -1
histogram = np.append(histogram, -1)
sign_change = ((np.roll(histogram, 1) - histogram) != 0).astype(int)
lines = np.where(sign_change == 1)[0]
distance = [lines[k] - lines[k - 1] for k in xrange(1, len(lines), 2)]
threshold_line_width = np.average(np.array(distance)) / 2
position = np.where(distance <= threshold_line_width)[0]
for p in xrange(len(position)):
lines = np.delete(lines, (position[p]) * 2 - p * 2)
lines = np.delete(lines, (position[p]) * 2 - p * 2)
if lines[0] == 0:
lines = np.delete(lines, 0)
lines = np.delete(lines, 0)
if lines[-1] == int(self.height):
lines = np.delete(lines, len(lines) - 1)
lines = np.delete(lines, len(lines) - 1)
text_lines.append(lines)
lines = self.get_same_line_position(text_lines)
empty_line = [0] * size
for l in lines:
croped_images = []
starts = l[0]
ends = l[1]
for i in xrange(len(starts)):
croped_images.append(self.image[starts[i]:ends[i], i * size:(i + 1) * size])
min_line = min(starts)
max_line = max(ends)
min_line_index = starts.index(min_line)
max_line_index = ends.index(max_line)
for k in xrange(self.NUMBER_OF_PART):
if k != min_line_index:
for z in xrange(starts[k] - min_line):
croped_images[k] = np.insert(croped_images[k], 0, empty_line, axis=0)
if k != max_line_index:
for z in xrange(max_line - ends[k]):
croped_images[k] = np.vstack([croped_images[k], empty_line])
temp_image = croped_images[0]
for ci in xrange(1, self.NUMBER_OF_PART):
temp_image = np.concatenate((temp_image, croped_images[ci]), axis=1)
# cv2.imwrite("bw_image_" + str(sys.argv[1].split("_")[2].split(".")[0]) + "_lines_" + str(self.line_index) + ".jpg", temp_image)
self.line_index += 1
self.detected_text_lines.append(temp_image)
self.get_characters()
input_filename = sys.argv[1]
# read input and covert to grayscale
img = cv2.imread(input_filename)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# invert image and thicken pixel lines
th, threshed = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY_INV|cv2.THRESH_OTSU)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
threshed = cv2.dilate(threshed, kernel, iterations=3)
threshed = cv2.erode(threshed, kernel, iterations=3)
# deskew image
rotated = straighten(threshed)
# find and draw the upper and lower boundary of each lines
hist = cv2.reduce(rotated, 1, cv2.REDUCE_AVG).reshape(-1)
th = 2
H, W = img.shape[:2]
uppers = [y for y in range(H-1) if hist[y]<=th and hist[y+1]>th]
lowers = [y for y in range(H-1) if hist[y]>th and hist[y+1]<=th]
for i in range(min(len(uppers), len(lowers))):
# isolate each line of text
line = threshed[uppers[i]-BUF:lowers[i]+BUF, 0:W].copy()
hist = cv2.reduce(line, 0, cv2.REDUCE_AVG).reshape(-1)
H, W = line.shape[:2]
lefts = [x for x in range(W-1) if gap_left(hist, x, th)]
rights = [x for x in range(W-1) if gap_right(hist, x, th, W)]
# http://stackoverflow.com/questions/34981144/split-text-lines-in-scanned-document/35014061
#
print cv2.__version__
img = cv2.imread('Afbeelding 2.png')
cv2.imshow('image',img)
#cv2.waitKey(0)
bin = (255 - img)
bin = cv2.cvtColor(bin,cv2.COLOR_BGR2GRAY)
pts = cv2.findNonZero(bin)
#box = cv2.minAreaRect(pts)
horProj = cv2.reduce(bin, 1, cv2.REDUCE_AVG)
horProj[horProj<=5.0] = 0
ycoords = []
y=0
count=0
isSpace=False
for i in xrange(0,horProj.size):
if (not isSpace):
if (not horProj[i]):
isSpace=True
count=1
y=i
else:
pylab.savefig(os.path.join(outdir, 'reduce_r.png'))
if 0:
print 'writing to %s' % outdir
img = cv2.imread(args.fn_in)
print type(img)
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
cv2.imwrite(os.path.join(outdir, 'reduce_01_gray.png'), gray)
print type(gray)
edges = cv2.Canny(gray, 125, 250, apertureSize=3)
print type(edges)
cv2.imwrite(os.path.join(outdir, 'reduce_02_edges.png'), edges)
#test = cv2.cvtColor(edges)
test = cv2.cv.GetMat(edges)
rowr = cv2.reduce(edges, 0, cv2.cv.CV_REDUCE_SUM)
colr = cv2.reduce(edges, 0, cv2.cv.CV_REDUCE_SUM)
matplotlib.pyplot.clf()
plt.subplot(211)
plt.plot(rowr)
plt.subplot(212)
plt.plot(colr)
pylab.savefig(os.path.join(outdir, 'reduce.png'))
matplotlib.pyplot.clf()
plt.plot(rowr)
pylab.savefig(os.path.join(outdir, 'reduce_r.png'))
matplotlib.pyplot.clf()
plt.plot(colr)
