def show_rotated(load_path, args=None, indices='all'):
'''
...
Args:
load_path:
args:
indices:
Returns:
'''
if args is None:
args = Arguments()
for i, file in enumerate(os.listdir(load_path)):
if file.endswith('.jpg') and (indices == 'all' or i in indices):
file_path = os.path.join(load_path, file)
img = processing.load_img(file_path)
components = find_blobs(img, args)
boxes = components.bboxes()
stencil = components.get_stencil()
angle, labels = correction_angle(components, args, False)
rotated_img = rotate(img, angle * 180 / np.pi)
plt.imshow(rotated_img)
plt.grid(True)
plt.show()
get_rotation_corrected_blobs(components, angle, labels, args)
def interface():
'''provides a terminal interface to execute the read function with an image path.
It prints out the result in the terminal and optionally extracts an IBAN.'''
path = input(
'please provide a path to an image containing text relative to this directory [example]\n> '
)
if path == '':
path = ('example.jpg')
while not os.path.exists(path):
path = input(
'This path doesn\'t exist. please provide a valid one [example]\n> '
)
args = Arguments()
img = load_img(path, args)
results = read(img, args)
out = ''
for line in results:
out += line + '\n'
print('\nFrom the provided image the following could be read:\n')
print(out)
get_iban = input('Do you want to extract an IBAN? y/n? [y] > ')
if get_iban == 'y' or get_iban == '':
ibans = detect_iban.find_iban(out)
if len(ibans) == 0:
print('No IBAN could be extracted.')
elif len(ibans) == 1:
copy = input(
'found the following IBAN: {} \n Do you want to copy it y/n? [y] > '
.format(ibans[0]))
if copy == '' or copy == 'y':
pyperclip.copy(ibans[0])
print('\nCopied {} to the clipboard'.format(ibans[0]))
else:
print('\nThe following IBANs were found:')
for i, iban in enumerate(ibans):
print(i, ':', iban)
idx = input(
'\nIf you want to copy any of these type its number > [0]')
if idx == '':
pyperclip.copy(ibans[0])
print('\nCopied {} to the clipboard'.format(ibans[0]))
elif int(idx) in range(len(ibans)):
pyperclip.copy(ibans[int(idx)])
print('\nCopied {} to the clipboard'.format(ibans[int(idx)]))
else:
print('\nThe input was not understood.')
print('\nexit')
def generate_char_data(load_path, args=None):
'''
...
'''
if args is None:
args = Arguments()
if args.method == 'threshold':
method = blob_extraction.find_blobs
elif args.method == 'mser':
method = mser_extraction.extract_mser
else:
assert False
# load alphabet
char_dict = args.char_dict
# load truth
with open(os.path.join(load_path, 'truth.txt')) as f:
content = f.readlines()
content = [x.strip() for x in content]
truth_blocks = []
current_block = []
for line in content:
if line == '':
truth_blocks.append(current_block)
current_block = []
else:
current_block.append(
np.array([char_dict[char] for char in line], dtype=np.int32))
# process images
char_images = []
char_truths = []
for i in range(args.n):
file = str(i) + '.jpg'
if file.endswith('.jpg'):
file_path = os.path.join(load_path, file)
img = processing.load_img(file_path, args)
components = method(img, args)
components = rotation_correct_and_line_order(components)
chars = components.extract(args)
# check if number of detected chars conicides with groundtruth
diff = sum(len(line) for line in truth_blocks[i]) - len(chars)
print(f'image {i}: #gt chars - #detected chars: {diff}')
if diff == 0:
char_images.append(chars)
char_truths.append(np.concatenate(truth_blocks[i]))
#print('image is valid')
return np.concatenate(char_images, axis=0), np.concatenate(char_truths,
axis=0)
def test_on_images(args):
'''
tests a trained model (specified by args.model_path) on the images labeled 1.jpg, 2.jpg.. in args.image_path
Args:
args: Arguments object
Returns:
predicted lines, ratio of correctly predicted lines,
ratio of correctly predicted lines not considering case,
ratio of lines with a correctly predicted amount of characters
'''
# get gt
with open(os.path.join(args.image_path, 'truth.txt')) as f:
content = f.readlines()
content = [x.strip() for x in content]
truth_blocks = []
current_block = []
for line in content:
if line == '':
truth_blocks.append(current_block)
current_block = []
else:
current_block.append(line)
total_lines = 0
correct_lines = 0
lowercase_correct_lines = 0
n_letters_correct_lines = 0
for i in range(args.n):
file = str(i) + '.jpg'
file_path = os.path.join(args.image_path, file)
img = load_img(file_path, args)
pred_lines = read(img, args)
gt_lines = truth_blocks[i]
print('prediction: ', pred_lines)
print('ground truth: ', gt_lines)
total_lines += len(gt_lines)
for i in range(min(len(gt_lines), len(pred_lines))):
correct_lines += (pred_lines[i] == gt_lines[i])
lowercase_correct_lines += (
pred_lines[i].lower() == gt_lines[i].lower())
n_letters_correct_lines += len(pred_lines[i]) == len(gt_lines[i])
print(
f'Out of {total_lines} lines, {correct_lines} ({100*correct_lines/total_lines:.2f}%) were predicted perfectly, '
f'and {lowercase_correct_lines} ({100*lowercase_correct_lines/total_lines:.2f}%) when neglecting case. '
f'in {n_letters_correct_lines} ({100*n_letters_correct_lines/total_lines:.2f}%) cases, at least the right number of letters was predicted.'
)
return total_lines, correct_lines / total_lines, lowercase_correct_lines / total_lines, n_letters_correct_lines / total_lines
def load_raws(self):
'''
loads raw images from the data directory and stores it as ndarrays
Returns:
list of ndarrays
'''
raws = []
for file in os.listdir(self.path):
if file.endswith('.jpg'):
file_path = os.path.join(self.path, file)
raws.append(processing.load_img(file_path))
return raws
lines.append(img[box[0]:box[1], box[2]:box[3]])
# bounding boxes in shape [x_min, y_min, width, height]
bboxes = [[box[2], box[0], box[3] - box[2], box[1] - box[0]]
for box in boxes]
return lines, bboxes
if __name__ == "__main__":
args = Arguments()
args.n_angles = 50
args.n_bins = 30
img = load_img('data/5.jpg')
rotated = correct_rot(img, args)
lines, bboxes = extract_lines(rotated, args)
plt.imshow(img[0:150, 0:200], cmap='gray')
plt.xticks([])
plt.yticks([])
#plt.savefig('plots/to_be_rotated.pdf', bbox_inches='tight')
plt.show()
o = 50
fig = plt.figure()
ax = fig.add_subplot(111)
ax.imshow(rotated[o:200, o:250], cmap='gray')
ax.grid(True)
angle = angles[np.argmin(H)]
angle = -angle * 180 / np.pi
# rotating image
img = rotate(img, angle)
return img
if __name__ == "__main__":
args = Arguments()
args.n_angles = 50
args.n_bins = 100
img = load_img('data_test/1.jpg')
rotated = correct_rot(img, args)
plt.imshow(img[0:150, 0:200], cmap='gray')
plt.xticks([])
plt.yticks([])
#plt.savefig('plots/to_be_rotated.pdf', bbox_inches='tight')
plt.show()
plt.imshow(rotated[50:200, 50:250], cmap='gray')
plt.grid(True)
plt.xticks([])
plt.yticks([])
#plt.savefig('plots/rotated.pdf', bbox_inches='tight')
plt.show()
tests the model on different font types and saves the result
Args:
args: Arguments object
prefix: folder in which folders of test images are saved
font_files: list of fonts to use
output_file: file to save results
'''
if font_files is None:
font_files = data_gen.get_font_files()
f = open(output_file, 'w')
for font in font_files:
print(ntpath.basename(font))
args.image_path = os.path.join(prefix, ntpath.basename(font))
total_lines, correct, correct_case_insensitive, correct_n_chars = test_on_images(
args)
f.write(
str(total_lines) + ' ' + str(correct) + ' ' +
str(correct_case_insensitive) + ' ' + str(correct_n_chars) + ' ' +
ntpath.basename(font) + '\n')
f.close()
if __name__ == '__main__':
args = Arguments()
args.documentation = True
args.method = 'mser'
img = load_img('IBANs/cropped/IMG_0892.jpg', args)
print(read(img, args))
if len(pixels) == 0:
continue
y_min = np.min(pixels[:, 1])
y_max = np.max(pixels[:, 1])
x_min = np.min(pixels[:, 0])
x_max = np.max(pixels[:, 0])
#heappush(boxes, (x_min, x_max, y_min, y_max))
boxes.append((x_min, x_max, y_min, y_max))
return np.array(boxes, dtype=np.int32)
if __name__ == "__main__":
args = Arguments()
img = processing.load_img('data_test/1.jpg')
scale = 1
orig = rescale(deepcopy(img), scale)
img = rescale(img, scale)
components = find_blobs(img, args)
boxes = components.bboxes()
stencil = components.get_stencil()
print()
fig = plt.figure()
ax = fig.add_subplot(111)
ax.imshow(orig, cmap='gray')
for box in boxes:
rect = Rectangle((box[2], box[0]),
box[3] - box[2],