基于LSTM和CTCLoss训练不定长图片验证码
Github项目地址:https://github.com/JansonJo/captcha_ocr.git
# coding=utf-8
"""
将三通道的图片转为灰度图进行训练
"""
import itertools
import os
import re
import random
import string
from collections import Counter
from os.path import join
import yaml
import cv2
import numpy as np
import tensorflow as tf
from keras import backend as K
from keras.callbacks import ModelCheckpoint, EarlyStopping, Callback
from keras.layers import Input, Dense, Activation, Dropout, BatchNormalization, Reshape, Lambda
from keras.layers.convolutional import Conv2D, MaxPooling2D
from keras.layers.merge import add, concatenate
from keras.layers.recurrent import GRU
from keras.models import Model, load_model
f = open('./config/config_demo.yaml', 'r', encoding='utf-8')
cfg = f.read()
cfg_dict = yaml.load(cfg)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
# config.gpu_options.per_process_gpu_memory_fraction = cfg_dict['System']['GpuMemoryFraction']
session = tf.Session(config=config)
K.set_session(session)
# System config
TRAIN_SET_PTAH = cfg_dict['System']['TrainSetPath']
VALID_SET_PATH = cfg_dict['System']['TestSetPath']
TEST_SET_PATH = cfg_dict['System']['TestSetPath']
MAX_TEXT_LEN = cfg_dict['System']['MaxTextLenth']
IMG_W = cfg_dict['System']['IMG_W']
IMG_H = cfg_dict['System']['IMG_H']
MODEL_NAME = cfg_dict['System']['ModelName']
LABEL_REGEX = cfg_dict['System']['LabelRegex']
ALPHABET = cfg_dict['System']['Alphabet']
# NeuralNet config
RNN_SIZE = cfg_dict['NeuralNet']['RNNSize']
DROPOUT = cfg_dict['NeuralNet']['Dropout']
# TrainParam config
MONITOR = cfg_dict['TrainParam']['EarlyStoping']['monitor']
PATIENCE = cfg_dict['TrainParam']['EarlyStoping']['patience']
MODE = cfg_dict['TrainParam']['EarlyStoping']['mode']
BASELINE = cfg_dict['TrainParam']['EarlyStoping']['baseline']
EPOCHS = cfg_dict['TrainParam']['Epochs']
BATCH_SIZE = cfg_dict['TrainParam']['BatchSize']
TEST_BATCH_SIZE = cfg_dict['TrainParam']['TestBatchSize']
TEST_SET_NUM = cfg_dict['TrainParam']['TestSetNum']
def get_counter(dirpath):
letters = ''
lens = []
for root, dirs, files in os.walk(dirpath):
for filename in files:
m = re.search(LABEL_REGEX, filename, re.M | re.I)
description = m.group(1)
lens.append(len(description))
letters += description
print('Max plate length in "%s":' % dirpath, max(Counter(lens).keys()))
return Counter(letters)
c_val = get_counter(VALID_SET_PATH)
c_train = get_counter(TRAIN_SET_PTAH)
letters_train = set(c_train.keys())
letters_val = set(c_val.keys())
print('letters_train: %s' % ''.join(sorted(letters_train)))
print('letters_val: %s' % ''.join(sorted(letters_val)))
if letters_train == letters_val:
print('Letters in train and val do match')
else:
raise Exception('Letters in train and val don\'t match')
# print(len(letters_train), len(letters_val), len(letters_val | letters_train))
# letters = sorted(list(letters_train))
# letters = '0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz'
letters = ALPHABET
if len(letters) == 0:
letters = string.digits + string.ascii_uppercase + string.ascii_lowercase
class_num = len(letters) + 1 # plus 1 for blank
print('Alphabet Letters:', ''.join(letters))
# Input data generator
def labels_to_text(labels):
# return ''.join(list(map(lambda x: letters[int(x)], labels)))
return ''.join([letters[int(x)] if int(x) != len(letters) else '' for x in labels])
def text_to_labels(text):
# return list(map(lambda x: letters.index(x), text))
return [letters.find(x) if letters.find(x) > -1 else len(letters) for x in text]
def is_valid_str(s):
for ch in s:
if not ch in letters:
return False
return True
class TextImageGenerator:
def __init__(self,
dirpath,
tag,
img_w, img_h,
batch_size,
downsample_factor,
max_text_len=MAX_TEXT_LEN):
self.img_h = img_h
self.img_w = img_w
self.batch_size = batch_size
self.max_text_len = max_text_len
self.downsample_factor = downsample_factor
img_dirpath = dirpath
self.samples = []
for filename in os.listdir(img_dirpath):
name, ext = os.path.splitext(filename)
if ext in ['.png', '.jpg']:
img_filepath = join(img_dirpath, filename)
m = re.search(LABEL_REGEX, filename, re.M | re.I)
description = m.group(1)
if len(description) < MAX_TEXT_LEN:
description = description + '_' * (MAX_TEXT_LEN - len(description))
# if is_valid_str(description):
# self.samples.append([img_filepath, description])
self.samples.append([img_filepath, description])
self.n = len(self.samples)
self.indexes = list(range(self.n))
self.cur_index = 0
# build data:
self.imgs = np.zeros((self.n, self.img_h, self.img_w))
self.texts = []
for i, (img_filepath, text) in enumerate(self.samples):
img = cv2.imread(img_filepath)
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) # cv2默认是BGR模式
img = cv2.resize(img, (self.img_w, self.img_h))
img = img.astype(np.float32)
img /= 255
# width and height are backwards from typical Keras convention
# because width is the time dimension when it gets fed into the RNN
self.imgs[i, :, :] = img
self.texts.append(text)
@staticmethod
def get_output_size():
return len(letters) + 1
def next_sample(self):
self.cur_index += 1
if self.cur_index >= self.n:
self.cur_index = 0
random.shuffle(self.indexes)
return self.imgs[self.indexes[self.cur_index]], self.texts[self.indexes[self.cur_index]]
def next_batch(self):
while True:
# width and height are backwards from typical Keras convention
# because width is the time dimension when it gets fed into the RNN
if K.image_data_format() == 'channels_first':
X_data = np.ones([self.batch_size, 1, self.img_w, self.img_h])
else:
X_data = np.ones([self.batch_size, self.img_w, self.img_h, 1])
Y_data = np.ones([self.batch_size, self.max_text_len])
input_length = np.ones((self.batch_size, 1)) * (self.img_w // self.downsample_factor - 2)
label_length = np.zeros((self.batch_size, 1))
source_str = []
for i in range(self.batch_size):
img, text = self.next_sample()
img = img.T
if K.image_data_format() == 'channels_first':
img = np.expand_dims(img, 0)
else:
img = np.expand_dims(img, -1)
X_data[i] = img
Y_data[i] = text_to_labels(text)
source_str.append(text)
text = text.replace("_", "") # important step
label_length[i] = len(text)
inputs = {
'the_input': X_data,
'the_labels': Y_data,
'input_length': input_length,
'label_length': label_length,
# 'source_str': source_str
}
outputs = {'ctc': np.zeros([self.batch_size])}
yield (inputs, outputs)
tiger = TextImageGenerator(VALID_SET_PATH, 'val', IMG_W, IMG_H, 8, 4)
for inp, out in tiger.next_batch():
print('Text generator output (data which will be fed into the neutral network):')
print('1) the_input (image)')
if K.image_data_format() == 'channels_first':
img = inp['the_input'][0, 0, :, :]
else:
img = inp['the_input'][0, :, :, 0]
# plt.imshow(img.T, cmap='gray')
# plt.show()
print('2) the_labels (plate number): %s is encoded as %s' %
(labels_to_text(inp['the_labels'][0]), list(map(int, inp['the_labels'][0]))))
print('3) input_length (width of image that is fed to the loss function): %d == %d / 4 - 2' %
(inp['input_length'][0], tiger.img_w))
print('4) label_length (length of plate number): %d' % inp['label_length'][0])
break
# # Loss and train functions, network architecture
def ctc_lambda_func(args):
y_pred, labels, input_length, label_length = args
# the 2 is critical here since the first couple outputs of the RNN
# tend to be garbage:
y_pred = y_pred[:, 2:, :]
return K.ctc_batch_cost(labels, y_pred, input_length, label_length)
downsample_factor = 4
def train(img_w=IMG_W, img_h=IMG_H, dropout=DROPOUT, batch_size=BATCH_SIZE, rnn_size=RNN_SIZE):
# Input Parameters
# Network parameters
conv_filters = 16
kernel_size = (3, 3)
pool_size = 2
time_dense_size = 32
if K.image_data_format() == 'channels_first':
input_shape = (1, img_w, img_h)
else:
input_shape = (img_w, img_h, 1)
global downsample_factor
downsample_factor = pool_size ** 2
tiger_train = TextImageGenerator(TRAIN_SET_PTAH, 'train', img_w, img_h, batch_size, downsample_factor)
tiger_val = TextImageGenerator(VALID_SET_PATH, 'val', img_w, img_h, batch_size, downsample_factor)
act = 'relu'
input_data = Input(name='the_input', shape=input_shape, dtype='float32')
inner = Conv2D(conv_filters, kernel_size, padding='same',
activation=None, kernel_initializer='he_normal',
name='conv1')(input_data)
inner = BatchNormalization()(inner) # add BN
inner = Activation(act)(inner)
inner = MaxPooling2D(pool_size=(pool_size, pool_size), name='max1')(inner)
inner = Conv2D(conv_filters, kernel_size, padding='same',
activation=None, kernel_initializer='he_normal',
name='conv2')(inner)
inner = BatchNormalization()(inner) # add BN
inner = Activation(act)(inner)
inner = MaxPooling2D(pool_size=(pool_size, pool_size), name='max2')(inner)
conv_to_rnn_dims = (img_w // (pool_size ** 2), (img_h // (pool_size ** 2)) * conv_filters)
inner = Reshape(target_shape=conv_to_rnn_dims, name='reshape')(inner)
# cuts down input size going into RNN:
inner = Dense(time_dense_size, activation=None, name='dense1')(inner)
inner = BatchNormalization()(inner) # add BN
inner = Activation(act)(inner)
if dropout:
inner = Dropout(dropout)(inner) # 防止过拟合
# Two layers of bidirecitonal GRUs
# GRU seems to work as well, if not better than LSTM:
gru_1 = GRU(rnn_size, return_sequences=True, kernel_initializer='he_normal', name='gru1')(inner)
gru_1b = GRU(rnn_size, return_sequences=True, go_backwards=True, kernel_initializer='he_normal', name='gru1_b')(
inner)
gru1_merged = add([gru_1, gru_1b])
gru_2 = GRU(rnn_size, return_sequences=True, kernel_initializer='he_normal', name='gru2')(gru1_merged)
gru_2b = GRU(rnn_size, return_sequences=True, go_backwards=True, kernel_initializer='he_normal', name='gru2_b')(
gru1_merged)
inner = concatenate([gru_2, gru_2b])
if dropout:
inner = Dropout(dropout)(inner) # 防止过拟合
# transforms RNN output to character activations:
inner = Dense(tiger_train.get_output_size(), kernel_initializer='he_normal',
name='dense2')(inner)
y_pred = Activation('softmax', name='softmax')(inner)
base_model = Model(inputs=input_data, outputs=y_pred)
base_model.summary()
labels = Input(name='the_labels', shape=[tiger_train.max_text_len], dtype='float32')
input_length = Input(name='input_length', shape=[1], dtype='int64')
label_length = Input(name='label_length', shape=[1], dtype='int64')
# Keras doesn't currently support loss funcs with extra parameters
# so CTC loss is implemented in a lambda layer
loss_out = Lambda(ctc_lambda_func, output_shape=(1,), name='ctc')([y_pred, labels, input_length, label_length])
model = Model(inputs=[input_data, labels, input_length, label_length], outputs=loss_out)
# the loss calc occurs elsewhere, so use a dummy lambda func for the loss
model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer='adadelta')
# if not load:
# captures output of softmax so we can decode the output during visualization
# test_func = K.function([input_data], [y_pred])
earlystoping = EarlyStopping(monitor=MONITOR, patience=PATIENCE, verbose=1, mode=MODE, baseline=BASELINE)
train_model_path = './tmp/train_' + MODEL_NAME
checkpointer = ModelCheckpoint(filepath=train_model_path,
verbose=1,
save_best_only=True)
if os.path.exists(train_model_path):
model.load_weights(train_model_path)
print('load model weights:%s' % train_model_path)
evaluator = Evaluate(model)
model.fit_generator(generator=tiger_train.next_batch(),
steps_per_epoch=tiger_train.n,
epochs=EPOCHS,
initial_epoch=1,
validation_data=tiger_val.next_batch(),
validation_steps=tiger_val.n,
callbacks=[checkpointer, earlystoping, evaluator])
base_model.save('./model/' + MODEL_NAME)
print('----train end----')
# For a real OCR application, this should be beam search with a dictionary
# and language model. For this example, best path is sufficient.
def decode_batch(out):
ret = []
for j in range(out.shape[0]):
out_best = list(np.argmax(out[j, 2:], 1))
out_best = [k for k, g in itertools.groupby(out_best)]
outstr = ''
for c in out_best:
if c < len(letters):
outstr += letters[c]
ret.append(outstr)
return ret
class Evaluate(Callback):
def __init__(self, model):
self.accs = []
self.model = model
def on_epoch_end(self, epoch, logs=None):
acc = evaluate(self.model)
self.accs.append(acc)
# Test on validation images
def evaluate(model):
global downsample_factor
tiger_test = TextImageGenerator(TEST_SET_PATH, 'test', IMG_W, IMG_H, TEST_BATCH_SIZE, downsample_factor)
net_inp = model.get_layer(name='the_input').input
net_out = model.get_layer(name='softmax').output
predict_model = Model(inputs=net_inp, outputs=net_out)
equalsIgnoreCaseNum = 0.00
equalsNum = 0.00
totalNum = 0.00
for inp_value, _ in tiger_test.next_batch():
batch_size = inp_value['the_input'].shape[0]
X_data = inp_value['the_input']
# net_out_value = sess.run(net_out, feed_dict={net_inp: X_data})
net_out_value = predict_model.predict(X_data)
pred_texts = decode_batch(net_out_value)
labels = inp_value['the_labels']
texts = []
for label in labels:
text = labels_to_text(label)
texts.append(text)
for i in range(batch_size):
# print('Predict: %s ---> Label: %s' % (pred_texts[i], texts[i]))
totalNum += 1
if pred_texts[i] == texts[i]:
equalsNum += 1
if pred_texts[i].lower() == texts[i].lower():
equalsIgnoreCaseNum += 1
else:
print('Predict: %s ---> Label: %s' % (pred_texts[i], texts[i]))
if totalNum >= TEST_SET_NUM:
break
print('---Result---')
print('Test num: %d, accuracy: %.5f, ignoreCase accuracy: %.5f' % (
totalNum, equalsNum / totalNum, equalsIgnoreCaseNum / totalNum))
return equalsIgnoreCaseNum / totalNum
if __name__ == '__main__':
train()