def textDetect(img, original):
""" Text detection using contours """
# Resize image
small = resize(img, 2000)
image = resize(original, 2000)
# Finding contours
mask = np.zeros(small.shape, np.uint8)
im2, cnt, hierarchy = cv2.findContours(np.copy(small), cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
#implt(img, 'gray')
# Variables for contour index and words' bounding boxes
index = 0
boundingBoxes = np.array([0,0,0,0])
bBoxes = []
# CCOMP hierarchy: [Next, Previous, First Child, Parent]
# cv2.RETR_CCOMP - contours into 2 levels
# Go through all contours in first level
while (index >= 0):
x,y,w,h = cv2.boundingRect(cnt[index])
# Get only the contour
cv2.drawContours(mask, cnt, index, (255, 255, 255), cv2.FILLED)
maskROI = mask[y:y+h, x:x+w]
# Ratio of white pixels to area of bounding rectangle
r = float(cv2.countNonZero(maskROI)) / (w * h)
# Limits for text (white pixel ratio, width, height)
# TODO Test h/w and w/h ratios
if r > 0.1 and 1600 > w > 10 and 1600 > h > 10 and h/w < 3 and w/h < 10 and (60 // h) * w < 1000:
bBoxes += [[x, y, w, h]]
# Index of next contour
index = hierarchy[0][index][0]
# Group intersecting rectangles
bBoxes = groupRectangles(bBoxes)
i = 0
f = open("output/words/normal/bounding_boxes_normal.txt","w")
for (x, y, w, h) in bBoxes:
boundingBoxes = np.vstack((boundingBoxes, np.array([x, y, x+w, y+h])))
cv2.imwrite("output/words/normal/"+str(i)+".jpg",image[y:y+h, x:x+w])
f.write(str(i) + "\t => \t" + "("+str(x)+","+str(y)+")"+","+"("+str(x+w)+","+str(y+h)+")"+"\n")
# cv2.rectangle(image, (x, y),(x+w,y+h), (0, 255, 0), 2)
i = i+1
#implt(image, t='Bounding rectangles')
# Recalculate coordinates to original scale
bBoxes = boundingBoxes.dot(ratio(image, small.shape[0])).astype(np.int64)
return bBoxes[1:]
def text_detect(img, original):
""" Text detection using contours """
# Resize image
small = resize(img, 2000)
image = resize(original, 2000)
# Finding contours
mask = np.zeros(small.shape, np.uint8)
cnt, hierarchy = cv2.findContours(np.copy(small), cv2.RETR_CCOMP,
cv2.CHAIN_APPROX_SIMPLE)
implt(img, 'gray')
# Variables for contour index and words' bounding boxes
index = 0
boxes = []
# CCOMP hierarchy: [Next, Previous, First Child, Parent]
# cv2.RETR_CCOMP - contours into 2 levels
# Go through all contours in first level
while (index >= 0):
x, y, w, h = cv2.boundingRect(cnt[index])
# Get only the contour
cv2.drawContours(mask, cnt, index, (255, 255, 255), cv2.FILLED)
maskROI = mask[y:y + h, x:x + w]
# Ratio of white pixels to area of bounding rectangle
r = cv2.countNonZero(maskROI) / (w * h)
# Limits for text (white pixel ratio, width, height)
# TODO Test h/w and w/h ratios
if r > 0.1 and 2000 > w > 10 and 1600 > h > 10 and h / w < 3 and w / h < 10:
boxes += [[x, y, w, h]]
# Index of next contour
index = hierarchy[0][index][0]
# Group intersecting rectangles
boxes = group_rectangles(boxes)
bounding_boxes = np.array([0, 0, 0, 0])
for (x, y, w, h) in boxes:
cv2.rectangle(image, (x, y), (x + w, y + h), (0, 255, 0), 8)
print((x, y, w, h))
bounding_boxes = np.vstack(
(bounding_boxes, np.array([x, y, x + w, y + h])))
implt(image, t='Bounding rectangles')
# Recalculate coordinates to original scale
boxes = bounding_boxes.dot(ratio(image, small.shape[0])).astype(np.int64)
return boxes[1:]
def edges_det(img, min_val, max_val):
""" Preprocessing (gray, thresh, filter, border) + Canny edge detection """
kernel = np.ones((2, 2), np.uint8)
img = cv2.cvtColor(resize(img), cv2.COLOR_BGR2GRAY)
# img = cv2.bilateralFilter(img, 9, 75, 75)
img = cv2.GaussianBlur(img, (5, 5), 0)
# img = cv2.GaussianBlur(img, (5, 5), 0)
implt(img, 'gray', 'blur')
img = cv2.adaptiveThreshold(img, 175, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 115, 1)
implt(img, 'gray', 'Adaptive Threshold')
img = cv2.erode(img, kernel, iterations=1)
implt(img, 'gray', 'erode')
img = cv2.dilate(img, kernel, iterations=1)
implt(img, 'gray', 'dilate')
img = cv2.medianBlur(img, 1)
# img = cv2.blur(img,(10,10))
img = cv2.copyMakeBorder(img,
5,
5,
5,
5,
cv2.BORDER_CONSTANT,
value=[0, 0, 0])
implt(img, 'gray', 'Median Blur + Border')
return cv2.Canny(img, min_val, max_val)
def edges_det(img, min_val, max_val):
""" Preprocessing (gray, thresh, filter, border) + Canny edge detection """
img = cv2.cvtColor(resize(img), cv2.COLOR_BGR2GRAY)
# Applying blur and threshold
img = cv2.bilateralFilter(img, 9, 75, 75)
img = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 115, 4)
implt(img, 'gray', 'Adaptive Threshold')
# Median blur replace center pixel by median of pixels under kelner
# => removes thin details
img = cv2.medianBlur(img, 11)
# Add black border - detection of border touching pages
# Contour can't touch side of image
img = cv2.copyMakeBorder(img,
5,
5,
5,
5,
cv2.BORDER_CONSTANT,
value=[0, 0, 0])
implt(img, 'gray', 'Median Blur + Border')
return cv2.Canny(img, min_val, max_val)
def edges_det(img, min_val, max_val):
img = cv2.cvtColor(resize(img), cv2.COLOR_BGR2GRAY)
img = cv2.bilateralFilter(img, 9, 75, 75)
img = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 115, 4)
implt(img, 'gray', 'Adaptive Threshold')
img = cv2.medianBlur(img, 11)
img = cv2.copyMakeBorder(img, 5, 5, 5, 5, cv2.BORDER_CONSTANT, value =[0, 0, 0])
implt(img, 'gray', 'Median Blur + Border')
return cv2.Canny(img, min_val, max_val)
def textDetectWatershed(thresh, original):
""" Text detection using watershed algorithm """
# According to: http://docs.opencv.org/trunk/d3/db4/tutorial_py_watershed.html
img = resize(original, 3000)
thresh = resize(thresh, 3000)
# noise removal
kernel = np.ones((3,3),np.uint8)
opening = cv2.morphologyEx(thresh,cv2.MORPH_OPEN,kernel, iterations = 3)
# sure background area
sure_bg = cv2.dilate(opening,kernel,iterations=3)
# Finding sure foreground area
dist_transform = cv2.distanceTransform(opening,cv2.DIST_L2,5)
ret, sure_fg = cv2.threshold(dist_transform,0.01*dist_transform.max(),255,0)
# Finding unknown region
sure_fg = np.uint8(sure_fg)
# cv2.imshow("image",sure_fg)
# cv2.waitKey(10)
unknown = cv2.subtract(sure_bg,sure_fg)
# Marker labelling
ret, markers = cv2.connectedComponents(sure_fg)
# Add one to all labels so that sure background is not 0, but 1
markers += 1
# Now, mark the region of unknown with zero
markers[unknown == 255] = 0
markers = cv2.watershed(img, markers)
#implt(markers, t='Markers')
image = img.copy()
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# Creating result array
boundingBoxes = np.array([0,0,0,0])
bBoxes = []
for mark in np.unique(markers):
# mark == 0 --> background
if mark == 0:
continue
# Draw it on mask and detect biggest contour
mask = np.zeros(gray.shape, dtype="uint8")
mask[markers == mark] = 255
cnts = cv2.findContours(mask.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[-2]
c = max(cnts, key=cv2.contourArea)
# Draw a bounding rectangle if it contains text
x,y,w,h = cv2.boundingRect(c)
cv2.drawContours(mask, c, 0, (255, 255, 255), cv2.FILLED)
maskROI = mask[y:y+h, x:x+w]
# Ratio of white pixels to area of bounding rectangle
r = float(cv2.countNonZero(maskROI)) / (w * h)
# Limits for text
# WORK ON THIS
if r > 0.1 and 2000 > w > 15 and 1500 > h > 15:
bBoxes += [[x, y, w, h]]
# Group intersecting rectangles
#bBoxes = groupRectangles(bBoxes)
i = 0
f = open("./output/words/watershed/bounding_boxes_watershed.txt","w")
for (x, y, w, h) in bBoxes:
boundingBoxes = np.vstack((boundingBoxes, np.array([x, y, x+w, y+h])))
cv2.imwrite("./output/words/watershed/"+str(i)+".jpg",image[y:y+h, x:x+w])
f.write(str(i) + "\t => \t" + "("+str(x)+","+str(y)+")"+","+"("+str(x+w)+","+str(y+h)+")"+"\n")
# cv2.rectangle(image, (x, y),(x+w,y+h), (0, 255, 0), 2)
i = i+1
#implt(image)
# Recalculate coordinates to original size
bBoxes = boundingBoxes.dot(ratio(original, 3000)).astype(np.int64)
return bBoxes[1:]
max_area = MIN_COUNTOUR_AREA
page_contour = np.array([[0, 0], [0, height - 5], [width - 5, height - 5],
[width - 5, 0]])
for cnt in contours:
perimeter = cv2.arcLength(cnt, True)
approx = cv2.approxPolyDP(cnt, 0.03 * perimeter, True)
# Page has 4 corners and it is convex
if (len(approx) == 4 and cv2.isContourConvex(approx)
and max_area < cv2.contourArea(approx) < MAX_COUNTOUR_AREA):
max_area = cv2.contourArea(approx)
page_contour = approx[:, 0]
# Sort corners and offset them
page_contour = four_corners_sort(page_contour)
return contour_offset(page_contour, (-5, -5))
page_contour = find_page_contours(edges_image, resize(image))
print("PAGE CONTOUR:")
print(page_contour)
implt(cv2.drawContours(resize(image), [page_contour], -1, (0, 255, 0), 3))
#Write To File
text = pytesseract.image_to_string(image)
print(text)
write_file(text, image)
for cnt in contours:
perimeter = cv2.arcLength(cnt, True)
approx = cv2.approxPolyDP(cnt, 0.03 * perimeter, True)
if (len(approx) == 4 and cv2.isContourConvex(approx)
and maxArea < cv2.contourArea(approx) < MAX_COUNTOUR_AREA):
maxArea = cv2.contourArea(approx)
pageContour = approx[:, 0]
pageContour = fourCornersSort(pageContour)
return contourOffset(pageContour, (-5, -5))
pageContour = findPageContours(closedEdges, resize(image))
print("PAGE CONTOUR:")
print(pageContour)
implt(cv2.drawContours(resize(image), [pageContour], -1, (0, 255, 0), 3))
pageContour = pageContour.dot(ratio(image))
def edgesDet(img, minVal, maxVal):
img = cv2.cvtColor(resize(img), cv2.COLOR_BGR2GRAY)
img = cv2.bilateralFilter(img, 9, 75, 75)
img = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 115, 4)
implt(img, 'gray', 'Adaptive Threshold')
# Create target points
tPoints = np.array([[0, 0], [0, height], [width, height], [width, 0]],
np.float32)
# getPerspectiveTransform() needs float32
if sPoints.dtype != np.float32:
sPoints = sPoints.astype(np.float32)
M = cv2.getPerspectiveTransform(sPoints, tPoints)
return cv2.warpPerspective(img, M, (int(width), int(height)))
#--
image = cv2.cvtColor(cv2.imread("test1.jpeg"), cv2.COLOR_BGR2RGB)
# Edge detection ()
imageEdges = edgesDet(image, 200, 250)
# Close gaps between edges (double page clouse => rectangle kernel)
closedEdges = cv2.morphologyEx(imageEdges, cv2.MORPH_CLOSE, np.ones((5, 11)))
pageContour = findPageContours(closedEdges, resize(image))
# Recalculate to original scale
pageContour = pageContour.dot(ratio(image))
newImage = perspImageTransform(image, pageContour)
cv2.imwrite("2_1.jpg", cv2.cvtColor(newImage, cv2.COLOR_BGR2RGB))
def text_detect(img, original):
""" Text detection using contours """
# Resize image
small = resize(img, 2000)
image = resize(original, 2000)
# Finding contours
mask = np.zeros(small.shape, np.uint8)
im2, cnt, hierarchy = cv2.findContours(np.copy(small), cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
# implt(img, 'gray')
# Variables for contour index and words' bounding boxes
index = 0
boxes = []
test_digits = np.array(np.zeros((1,784)))
i = 0
# CCOMP hierarchy: [Next, Previous, First Child, Parent]
# cv2.RETR_CCOMP - contours into 2 levels
# Go through all contours in first level
while (index >= 0):
x,y,w,h = cv2.boundingRect(cnt[index])
# Get only the contour
cv2.drawContours(mask, cnt, index, (255, 255, 255), cv2.FILLED)
maskROI = mask[y:y+h, x:x+w]
# Ratio of white pixels to area of bounding rectangle
r = cv2.countNonZero(maskROI) / (w * h)
# Limits for text (white pixel ratio, width, height)
# TODO Test h/w and w/h ratios
# if r > 0.1 and 2000 > w > 10 and 1600 > h > 10 and h/w < 3 and w/h < 10:
if r > 0.1 and 40 > w > 1 and 1600 > h > 10 and (h/w < 3 or w/h < 10):
boxes += [[x, y, w, h]]
roi = image[y:y+h, x:x+w]
digit = cv2.cvtColor(roi, cv2.COLOR_RGB2GRAY)
digit, j = digit_preprocessing(digit)
cv2.imwrite("../data/form/output{0}.jpg".format(i), digit)
digit_row = np.resize(digit,(1,784))
test_digits = np.append(test_digits,digit_row,axis=0)
i += 1
# Index of next contour
index = hierarchy[0][index][0]
# Group intersecting rectangles
boxes = group_rectangles(boxes)
bounding_boxes = np.array([0,0,0,0])
for (x, y, w, h) in boxes:
cv2.rectangle(image, (x, y),(x+w,y+h), (0, 255, 0), 8)
bounding_boxes = np.vstack((bounding_boxes, np.array([x, y, x+w, y+h])))
implt(image, t='Bounding rectangles')
# Recalculate coordinates to original scale
boxes = bounding_boxes.dot(ratio(image, small.shape[0])).astype(np.int64)
return test_digits, boxes[1:]
