def main(argv):
# load image
print('loading image:', argv)
image = io.imread(argv, as_gray=True)
# split into letters
letters = split_letters(image)
probs = []
for i, letter in enumerate(letters):
letter = letter.astype('float32') / 255
predict_input_fn = tf.estimator.inputs.numpy_input_fn(x={'x': letter},
y=None,
num_epochs=1,
shuffle=False)
predictions = captcha_classifier.predict(input_fn=predict_input_fn)
for pred_dict in predictions:
class_id = pred_dict['classes']
probs.append(str(class_id))
print(''.join(probs))
def main(argv):
args = parser.parse_args(argv[1:])
if args.fname == None:
print('Capture Image is missing!')
return None
# load image
print('loading image:', args.fname)
image = io.imread(args.fname)
# split into letters
letters = split_letters(image)
if letters == None:
print('Letters is not valid')
return None
# predict
content = ''
probs = []
for i, letter in enumerate(letters):
letter = letter.astype('float32') / 255
predict_input_fn = tf.estimator.inputs.numpy_input_fn(x={'x': letter},
y=None,
num_epochs=1,
shuffle=False)
predictions = captcha_classifier.predict(input_fn=predict_input_fn)
for pred_dict in predictions:
class_id = pred_dict['classes']
prob = pred_dict['probabilities'][class_id]
class_name = chr(65 + class_id) # from 65 : 'A'
print('predict for {} char: `{}` with probability: {:.3f}%'.format(
i + 1, class_name, prob * 100))
content += class_name
probs.append(prob)
print('Captcha: `{}` with confident: `{:.3f}%`'.format(
content, 100 * sum(probs) / len(probs)))
accuracy = str(round(100 * sum(probs) / len(probs), 2))
print('ACCURACY-' + accuracy)
result = 'RESULT-' + content
print(result)
# load image content json file
with open(DATA_MAP) as f:
image_contents = json.load(f)
# load image and save letters
counter = 0
failCounter = 0
for fname, contents in image_contents.iteritems():
counter += 1
image = io.imread(os.path.join(DATA_FULL_DIR, fname))
# split image
letters = []
try:
letters = split_letters(image, debug=True)
except:
print('split_letters failed')
failCounter += 1
else:
if letters != None:
fname = fname.replace('.jpg', '.png')
print(counter)
for i, letter in enumerate(letters):
content = contents[i]
# add to dataset
data_x.append(letter)
data_y.append(np.uint8(ord(content) - 48)) # 48: '0'
# save letter into train folder
fpath = os.path.join(DATA_TRAIN_DIR, content)
# build image contents map
image_contents = build_data_map(DATA_FULL_DIR)
# load image and save letters
counter = 0
for fname, contents in image_contents.iteritems():
counter += 1
print(counter, fname, contents)
original_image = io.imread(os.path.join(DATA_FULL_DIR, fname))
grayscale_image = rgb2gray(original_image)
# split image
letters = split_letters(grayscale_image,
num_letters=len(contents),
debug=True)
if letters != None:
for i, letter in enumerate(letters):
content = contents[i]
# add to dataset
data_x.append(letter)
data_y.append(np.uint8(ord(content) - 65)) # 65: 'A'
# save letter into train folder
fpath = os.path.join(DATA_TRAIN_DIR, content)
if not os.path.exists(fpath):
os.makedirs(fpath)
letter_fname = os.path.join(
fpath,
str(i + 1) + '-' + strip_extension(fname) + '.png')