def show_differences(image, segments, ground_classes, result_classes):
image = image.copy()
good = (ground_classes == result_classes)
good.shape = (len(good), ) #transform nx1 matrix into vector
draw_segments(image, segments[good, :], (0, 255, 0))
draw_segments(image, segments[numpy.logical_not(good), :], (0, 0, 255))
show_image_and_wait_for_key(image, "differences")
def show_differences( image, segments, ground_classes, result_classes):
image= image.copy()
good= (ground_classes==result_classes)
good.shape= (len(good),) #transform nx1 matrix into vector
draw_segments( image, segments[good,:], (0,255,0) )
draw_segments( image, segments[numpy.logical_not(good),:], (0,0,255) )
show_image_and_wait_for_key(image, "differences")
def display(self, display_before=False):
copy= self.image.copy()
if display_before:
show_image_and_wait_for_key(copy, "image before segmentation")
copy.fill( (255,255,255) )
cv2.drawContours(copy, self.contours, contourIdx=-1, color=(0,0,0))
show_image_and_wait_for_key( copy, "ContourSegmenter contours")
copy= self.image.copy()
draw_segments( copy, self.segments)
show_image_and_wait_for_key(copy, "image after segmentation by "+self.__class__.__name__)
def display(self, display_before=False):
copy = self.image.copy()
if display_before:
show_image_and_wait_for_key(copy, "image before segmentation")
copy.fill((255, 255, 255))
cv2.drawContours(copy, self.contours, contourIdx=-1, color=(0, 0, 0))
show_image_and_wait_for_key(copy, "ContourSegmenter contours")
copy = self.image.copy()
draw_segments(copy, self.segments)
show_image_and_wait_for_key(
copy, "image after segmentation by " + self.__class__.__name__)
def display( self, display_before=False, image_override=None ):
'''shows the effect of this filter'''
if not image_override is None:
copy= image_override
else:
try:
copy= self.image.copy()
except AttributeError:
raise Exception("You need to set the Filter.image attribute for displaying")
copy= BrightnessProcessor(brightness=0.6).process( copy )
s, g= self._input, self.good_segments_indexes
draw_segments( copy, s[g], (0,255,0) )
draw_segments( copy, s[True-g], (0,0,255) )
show_image_and_wait_for_key( copy, "segments filtered by "+self.__class__.__name__)
def ground(self, imagefile, segments, _=None):
'''asks the user to label each segment as either a character or "<" for unknown'''
print '''For each shown segment, please write the character that it represents, or spacebar if it's not a character. To undo a classification, press backspace. Press ESC when completed, arrow keys to move'''
i = 0
if imagefile.isGrounded():
classes = classes_from_numpy(imagefile.ground.classes)
segments = imagefile.ground.segments
else:
classes = [BLANK_CLASS] * len(
segments
) #char(10) is newline. it represents a non-assigned label, and will b filtered
done = False
allowed_chars = map(
ord, string.digits + string.letters + string.punctuation)
while not done:
image = imagefile.image.copy()
draw_segments(image, [segments[i]])
draw_classes(image, segments, classes)
key = show_image_and_wait_for_key(image, "segment " + str(i))
if key == 27: #ESC
break
elif key == 8: #backspace
classes[i] = BLANK_CLASS
i += 1
elif key == 32: #space
classes[i] = NOT_A_SEGMENT
i += 1
elif key == 65361: #<-
i -= 1
elif key == 65363: #->
i += 1
elif key in allowed_chars:
classes[i] = unichr(key)
i += 1
if i >= len(classes):
i = 0
if i < 0:
i = len(classes) - 1
classes = numpy.array(classes)
is_segment = classes != NOT_A_SEGMENT
classes = classes[is_segment]
segments = segments[is_segment]
classes = list(classes)
classes = classes_to_numpy(classes)
print "classified ", numpy.count_nonzero(
classes != classes_to_numpy(BLANK_CLASS)
), "characters out of", max(classes.shape)
imagefile.set_ground(segments, classes)
def display(self, display_before=False, image_override=None):
'''shows the effect of this filter'''
if not image_override is None:
copy = image_override
else:
try:
copy = self.image.copy()
except AttributeError:
raise Exception(
"You need to set the Filter.image attribute for displaying"
)
copy = BrightnessProcessor(brightness=0.6).process(copy)
s, g = self._input, self.good_segments_indexes
draw_segments(copy, s[g], (0, 255, 0))
draw_segments(copy, s[True - g], (0, 0, 255))
show_image_and_wait_for_key(
copy, "segments filtered by " + self.__class__.__name__)
def ground( self, imagefile, segments, _=None ):
'''asks the user to label each segment as either a character or "<" for unknown'''
print '''For each shown segment, please write the character that it represents, or spacebar if it's not a character. To undo a classification, press backspace. Press ESC when completed, arrow keys to move'''
i=0
if imagefile.isGrounded():
classes= classes_from_numpy( imagefile.ground.classes)
segments= imagefile.ground.segments
else:
classes= [BLANK_CLASS]*len(segments) #char(10) is newline. it represents a non-assigned label, and will b filtered
done= False
allowed_chars= map( ord, string.digits+string.letters+string.punctuation )
while not done:
image= imagefile.image.copy()
draw_segments( image, [segments[ i ]])
draw_classes( image, segments, classes )
key= show_image_and_wait_for_key( image, "segment "+str(i))
if key==27: #ESC
break
elif key==8: #backspace
classes[i]= BLANK_CLASS
i+=1
elif key==32: #space
classes[i]= NOT_A_SEGMENT
i+=1
elif key==65361: #<-
i-=1
elif key==65363: #->
i+=1
elif key in allowed_chars:
classes[i]= unichr(key)
i+=1
if i>=len(classes):
i=0
if i<0:
i=len(classes)-1
classes= numpy.array( classes )
is_segment= classes != NOT_A_SEGMENT
classes= classes[ is_segment ]
segments= segments[ is_segment ]
classes= list(classes)
classes= classes_to_numpy( classes )
print "classified ",numpy.count_nonzero( classes != classes_to_numpy(BLANK_CLASS) ), "characters out of", max(classes.shape)
imagefile.set_ground( segments, classes )
def preprocess_with_display(image):
copy = image.copy()
cv2.imshow('Display', image)
cv2.waitKey(0)
image = cv2.GaussianBlur(image, (5, 5), 0) # Blurring the image
cv2.imshow('Display', image)
print "After Guassian blurring"
cv2.waitKey(0)
image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) # Grey scaling the image
cv2.imshow('Display', image) # Displaying the image
print "After Greyscaling"
cv2.waitKey(0)
image = cv2.adaptiveThreshold(
image,
255,
cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY,
11, 10)
''' Thresholding the image (Converts greyscale to binary),
using adaptive threshold for best results '''
cv2.imshow('Display', image)
im2, contours, hierarchy = cv2.findContours(
image,
cv2.RETR_LIST,
cv2.CHAIN_APPROX_SIMPLE)
''' Finding the contours in the image'''
image.fill(255)
cv2.drawContours(image, contours, -1, (0, 0, 255))
''' Drawing the contours on the image before displaying'''
cv2.imshow('Display', image)
print "After detecting Contours"
cv2.waitKey(0)
contours.reverse()
segments = segments_to_numpy([cv2.boundingRect(c) for c in contours])
segments = numpy.delete(segments, 0, 0)
euler_list, inner_segments = find_euler_and_inner_segments(segments, 1)
segments, euler_list, central_x, central_y = segment_blocks(
segments, inner_segments, euler_list)
draw_segments(copy, segments)
'''Draw the segments on the copy image (cant add color to greyscaled image)'''
cv2.imshow('Display', copy)
print "After Segmentation"
cv2.waitKey(0)
return image, segments, euler_list, central_x, central_y
