深度学习实战之CNN验证码识别案列

任务要求

我们使用卷积神经网络来实现验证码识别案列，具体流程如下：

1、使用python的captcha模块生成验证码图片。 2、使用tensorflow框架搭建神经网络模型。 3、将数据喂入搭建好的神经网络模型中。 4、保存训练好的网络模型。

下面我们来看具体的细节。 一、定义字符集，验证码一般为数字、字母。练习的时候可以先只考虑数字的情况，这样模型训练的会快些。代码如下：

number = ['0','1','2','3','4','5','6','7','8','9'] alphabet = ['a','b','c','d','e','f','g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v','w','x','y','z'] ALPHABET = ['A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P','Q','R','S','T','U','V','W','X','Y','Z']

二、下面我们要从给定的字符集中选择4个字符，生成160*60的验证码图片，并将图片转化为numpy数组。然后将选择的四个字符生成为词向量形式。 1、生成图片并转化为数组。

def random_captcha_text(char_set=number+alphabet+ALPHABET, captcha_size=4): captcha_text = [] for i in range(captcha_size): c = random.choice(char_set) captcha_text.append(c) return captcha_text def gen_captcha_text_and_image(): image = ImageCaptcha() captcha_text = random_captcha_text() captcha_text = ''.join(captcha_text) captcha = image.generate(captcha_text) #image.write(captcha_text, captcha_text + '.jpg') captcha_image = Image.open(captcha) captcha_image = captcha_image.convert('L') captcha_image = captcha_image.point(lambda i: 255 - i) #将图片取反，黑色变为白色，白色变为黑色，这样模型收敛更快 captcha_image = np.array(captcha_image) return captcha_text, captcha_image

2、传入验证码文本，转化为词向量的形式，假设我们现在只使用数字集0-9。那么就是10分类，我们用一个长度为10的向量来表示一个数字，比如[1, 0, 0, 0, 0, 0, 0, 0, 0, 0]表示数字0，[0, 1, 0, 0, 0, 0, 0, 0, 0, 0]表示数字1。我们有四个字符，所以是一个410的矩阵，再将这个矩阵拉平为一维的，就是长度为40的向量。 如果我们现在采用数字加大小写字母为字符集，那就是4(10+26+26)，再将矩阵拉平，就是长度为248的向量。代码如下：

def text2vec(text): text_len = len(text) if text_len > MAX_CAPTCHA: raise ValueError('验证码最长4个字符') vector = np.zeros(MAX_CAPTCHA*CHAR_SET_LEN) def char2pos(c): if c =='_': k = 62 return k k = ord(c)-48 if k > 9: k = ord(c) - 55 if k > 35: k = ord(c) - 61 if k > 61: raise ValueError('No Map') return k for i, c in enumerate(text): idx = i * CHAR_SET_LEN + char2pos(c) vector[idx] = 1 return vector

三、以上代码每次只生成一张验证码，当然每次传入网络一个样本也可以，但我们习惯一次喂入多个样本，所以我们还要一次性生成多张图片传入网络。代码如下。

# 生成一个训练batch def get_next_batch(batch_size=128): batch_x = np.zeros([batch_size, IMAGE_HEIGHT*IMAGE_WIDTH]) batch_y = np.zeros([batch_size, MAX_CAPTCHA*CHAR_SET_LEN]) # 有时生成图像大小不是(60, 160, 3) def wrap_gen_captcha_text_and_image(): while True: text, image = gen_captcha_text_and_image() if image.shape == (60, 160, 3): return text, image for i in range(batch_size): text, image = wrap_gen_captcha_text_and_image() image = convert2gray(image) batch_x[i,:] = image.flatten() / 255 # (image.flatten()-128)/128 mean为0 batch_y[i,:] = text2vec(text) return batch_x, batch_y def convert2gray(img): if len(img.shape) > 2: gray = np.mean(img, -1) # 上面的转法较快，正规转法如下 # r, g, b = img[:,:,0], img[:,:,1], img[:,:,2] # gray = 0.2989 * r + 0.5870 * g + 0.1140 * b return gray else: return img

四、现在图片生成好了，对应的词向量也生成好了，要开始搭建网络了，我们采用三层卷积，一层全连接层，最后输出成，代码如下：

# 定义CNN def crack_captcha_cnn(w_alpha=0.01, b_alpha=0.1): x = tf.reshape(X, shape=[-1, IMAGE_HEIGHT, IMAGE_WIDTH, 1]) #w_c1_alpha = np.sqrt(2.0/(IMAGE_HEIGHT*IMAGE_WIDTH)) # #w_c2_alpha = np.sqrt(2.0/(3*3*32)) #w_c3_alpha = np.sqrt(2.0/(3*3*64)) #w_d1_alpha = np.sqrt(2.0/(8*32*64)) #out_alpha = np.sqrt(2.0/1024) # 3 conv layer w_c1 = tf.Variable(w_alpha*tf.random_normal([3, 3, 1, 32])) b_c1 = tf.Variable(b_alpha*tf.random_normal([32])) conv1 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(x, w_c1, strides=[1, 1, 1, 1], padding='SAME'), b_c1)) conv1 = tf.nn.max_pool(conv1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') conv1 = tf.nn.dropout(conv1, keep_prob) w_c2 = tf.Variable(w_alpha*tf.random_normal([3, 3, 32, 64])) b_c2 = tf.Variable(b_alpha*tf.random_normal([64])) conv2 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(conv1, w_c2, strides=[1, 1, 1, 1], padding='SAME'), b_c2)) conv2 = tf.nn.max_pool(conv2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') conv2 = tf.nn.dropout(conv2, keep_prob) w_c3 = tf.Variable(w_alpha*tf.random_normal([3, 3, 64, 64])) b_c3 = tf.Variable(b_alpha*tf.random_normal([64])) conv3 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(conv2, w_c3, strides=[1, 1, 1, 1], padding='SAME'), b_c3)) conv3 = tf.nn.max_pool(conv3, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') conv3 = tf.nn.dropout(conv3, keep_prob) # Fully connected layer w_d = tf.Variable(w_alpha*tf.random_normal([8*32*40, 1024])) b_d = tf.Variable(b_alpha*tf.random_normal([1024])) dense = tf.reshape(conv3, [-1, w_d.get_shape().as_list()[0]]) dense = tf.nn.relu(tf.add(tf.matmul(dense, w_d), b_d)) dense = tf.nn.dropout(dense, keep_prob) w_out = tf.Variable(w_alpha*tf.random_normal([1024, MAX_CAPTCHA*CHAR_SET_LEN])) b_out = tf.Variable(b_alpha*tf.random_normal([MAX_CAPTCHA*CHAR_SET_LEN])) out = tf.add(tf.matmul(dense, w_out), b_out) #out = tf.nn.softmax(out) return out

五、网络构建好了，现在需要构建损失函数，以及准确率等等，并开始训练了。具体代码如下：

# 训练 def train_crack_captcha_cnn(): output = crack_captcha_cnn() # loss #loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(output, Y)) loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(output, Y)) # 最后一层用来分类的softmax和sigmoid有什么不同？ # optimizer 为了加快训练 learning_rate应该开始大，然后慢慢衰 optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss) predict = tf.reshape(output, [-1, MAX_CAPTCHA, CHAR_SET_LEN]) max_idx_p = tf.argmax(predict, 2) max_idx_l = tf.argmax(tf.reshape(Y, [-1, MAX_CAPTCHA, CHAR_SET_LEN]), 2) correct_pred = tf.equal(max_idx_p, max_idx_l) accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32)) saver = tf.train.Saver() with tf.Session() as sess: sess.run(tf.global_variables_initializer()) step = 0 while True: batch_x, batch_y = get_next_batch(64) _, loss_ = sess.run([optimizer, loss], feed_dict={X: batch_x, Y: batch_y, keep_prob: 0.75}) print(step, loss_) # 每100 step计算一次准确率 if step % 10 == 0: batch_x_test, batch_y_test = get_next_batch(100) acc = sess.run(accuracy, feed_dict={X: batch_x_test, Y: batch_y_test, keep_prob: 1.}) print(step, acc) # 如果准确率大于50%,保存模型,完成训练 if acc > 0.50: saver.save(sess, "./model/crack_capcha.model", global_step=step) break step += 1

整个结构基本就是这样，如果只采用数字集的话，基本一千次迭代，半小时左右，准确率就能到90%以上。如果采用数字加大小写字母，时间会稍微久一点。下面是完整的代码：

import numpy as np import tensorflow as tf from captcha.image import ImageCaptcha import numpy as np import matplotlib.pyplot as plt from PIL import Image import random number = ['0','1','2','3','4','5','6','7','8','9'] #alphabet = ['a','b','c','d','e','f','g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v','w','x','y','z'] #ALPHABET = ['A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P','Q','R','S','T','U','V','W','X','Y','Z'] #def random_captcha_text(char_set=number+alphabet+ALPHABET, captcha_size=4): def random_captcha_text(char_set=number, captcha_size=4): captcha_text = [] for i in range(captcha_size): c = random.choice(char_set) captcha_text.append(c) return captcha_text def gen_captcha_text_and_image(): image = ImageCaptcha() captcha_text = random_captcha_text() captcha_text = ''.join(captcha_text) captcha = image.generate(captcha_text) #image.write(captcha_text, captcha_text + '.jpg') captcha_image = Image.open(captcha) captcha_image = np.array(captcha_image) return captcha_text, captcha_image def convert2gray(img): if len(img.shape) > 2: gray = np.mean(img, -1) # 上面的转法较快，正规转法如下 # r, g, b = img[:,:,0], img[:,:,1], img[:,:,2] # gray = 0.2989 * r + 0.5870 * g + 0.1140 * b return gray else: return img def text2vec(text): text_len = len(text) if text_len > MAX_CAPTCHA: raise ValueError('验证码最长4个字符') vector = np.zeros(MAX_CAPTCHA*CHAR_SET_LEN) def char2pos(c): if c =='_': k = 62 return k k = ord(c)-48 if k > 9: k = ord(c) - 55 if k > 35: k = ord(c) - 61 if k > 61: raise ValueError('No Map') return k for i, c in enumerate(text): idx = i * CHAR_SET_LEN + int(c) vector[idx] = 1 return vector # 传入验证码字符文本，生成对应的词向量 # def text2vec(text): # text_len = len(text) # if text_len > MAX_CAPTCHA: # raise ValueError('验证码最长4个字符') # vector = np.zeros(MAX_CAPTCHA*CHAR_SET_LEN) # def char2pos(c): # if c =='_': # k = 62 # return k # k = ord(c)-48 # if k > 9: # k = ord(c) - 55 # if k > 35: # k = ord(c) - 61 # if k > 61: # raise ValueError('No Map') # return k # for i, c in enumerate(text): # idx = i * CHAR_SET_LEN + char2pos(c) # vector[idx] = 1 # return vector # 向量转回文本 def vec2text(vec): """ char_pos = vec.nonzero()[0] text=[] for i, c in enumerate(char_pos): char_at_pos = i #c/63 char_idx = c % CHAR_SET_LEN if char_idx < 10: char_code = char_idx + ord('0') elif char_idx <36: char_code = char_idx - 10 + ord('A') elif char_idx < 62: char_code = char_idx- 36 + ord('a') elif char_idx == 62: char_code = ord('_') else: raise ValueError('error') text.append(chr(char_code)) """ text=[] char_pos = vec.nonzero()[0] for i, c in enumerate(char_pos): number = i % 10 text.append(str(number)) return "".join(text) """ #向量（大小MAX_CAPTCHA*CHAR_SET_LEN）用0,1编码 每63个编码一个字符，这样顺利有，字符也有 vec = text2vec("F5Sd") text = vec2text(vec) print(text) # F5Sd vec = text2vec("SFd5") text = vec2text(vec) print(text) # SFd5 """ # 生成一个训练batch def get_next_batch(batch_size=128): batch_x = np.zeros([batch_size, IMAGE_HEIGHT*IMAGE_WIDTH]) batch_y = np.zeros([batch_size, MAX_CAPTCHA*CHAR_SET_LEN]) # 有时生成图像大小不是(60, 160, 3) def wrap_gen_captcha_text_and_image(): while True: text, image = gen_captcha_text_and_image() if image.shape == (60, 160, 3): return text, image for i in range(batch_size): text, image = wrap_gen_captcha_text_and_image() image = convert2gray(image) #将二维数组拉成一维数组 batch_x[i,:] = image.flatten() / 255 # (image.flatten()-128)/128 mean为0 batch_y[i,:] = text2vec(text) return batch_x, batch_y # 定义CNN ，这里使用三层卷积和一层全连接操作，最后输出 def crack_captcha_cnn(w_alpha=0.01, b_alpha=0.1): x = tf.reshape(X, shape=[-1, IMAGE_HEIGHT, IMAGE_WIDTH, 1]) #w_c1_alpha = np.sqrt(2.0/(IMAGE_HEIGHT*IMAGE_WIDTH)) # #w_c2_alpha = np.sqrt(2.0/(3*3*32)) #w_c3_alpha = np.sqrt(2.0/(3*3*64)) #w_d1_alpha = np.sqrt(2.0/(8*32*64)) #out_alpha = np.sqrt(2.0/1024) # 3 conv layer w_c1 = tf.Variable(w_alpha*tf.random_normal([3, 3, 1, 32])) b_c1 = tf.Variable(b_alpha*tf.random_normal([32])) conv1 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(x, w_c1, strides=[1, 1, 1, 1], padding='SAME'), b_c1)) conv1 = tf.nn.max_pool(conv1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') conv1 = tf.nn.dropout(conv1, keep_prob) w_c2 = tf.Variable(w_alpha*tf.random_normal([3, 3, 32, 64])) b_c2 = tf.Variable(b_alpha*tf.random_normal([64])) conv2 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(conv1, w_c2, strides=[1, 1, 1, 1], padding='SAME'), b_c2)) conv2 = tf.nn.max_pool(conv2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') conv2 = tf.nn.dropout(conv2, keep_prob) w_c3 = tf.Variable(w_alpha*tf.random_normal([3, 3, 64, 64])) b_c3 = tf.Variable(b_alpha*tf.random_normal([64])) conv3 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(conv2, w_c3, strides=[1, 1, 1, 1], padding='SAME'), b_c3)) conv3 = tf.nn.max_pool(conv3, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') conv3 = tf.nn.dropout(conv3, keep_prob) # Fully connected layer w_d = tf.Variable(w_alpha*tf.random_normal([8*20*64, 1024])) b_d = tf.Variable(b_alpha*tf.random_normal([1024])) dense = tf.reshape(conv3, [-1, w_d.get_shape().as_list()[0]]) dense = tf.nn.relu(tf.add(tf.matmul(dense, w_d), b_d)) dense = tf.nn.dropout(dense, keep_prob) w_out = tf.Variable(w_alpha*tf.random_normal([1024, MAX_CAPTCHA*CHAR_SET_LEN])) b_out = tf.Variable(b_alpha*tf.random_normal([MAX_CAPTCHA*CHAR_SET_LEN])) out = tf.add(tf.matmul(dense, w_out), b_out) return out # 网络搭建好之后需要构建损失函数，以及准确率，并开始训练 def train_crack_captcha_cnn(): output = crack_captcha_cnn() loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(output, Y)) optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss) predict = tf.reshape(output, [-1, MAX_CAPTCHA, CHAR_SET_LEN]) max_idx_p = tf.argmax(predict, 2) max_idx_l = tf.argmax(tf.reshape(Y, [-1, MAX_CAPTCHA, CHAR_SET_LEN]), 2) correct_pred = tf.equal(max_idx_p, max_idx_l) accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32)) saver = tf.train.Saver() with tf.Session() as sess: sess.run(tf.global_variables_initializer()) step = 0 while True: batch_x, batch_y = get_next_batch(64) _, loss_ = sess.run([optimizer, loss], feed_dict={X: batch_x, Y: batch_y, keep_prob: 0.75}) print(step, loss_) # 每100 step计算一次准确率 if step % 10 == 0: batch_x_test, batch_y_test = get_next_batch(100) acc = sess.run(accuracy, feed_dict={X: batch_x_test, Y: batch_y_test, keep_prob: 1.}) print(step, acc) # 如果准确率大于50%,保存模型,完成训练 if acc > 0.50: saver.save(sess, "./model/crack_capcha.model", global_step=step) break step += 1 def crack_captcha(captcha_image): output = crack_captcha_cnn() saver = tf.train.Saver() with tf.Session() as sess: saver.restore(sess, "./model/crack_capcha.model-810") predict = tf.argmax(tf.reshape(output, [-1, MAX_CAPTCHA, CHAR_SET_LEN]), 2) text_list = sess.run(predict, feed_dict={X: [captcha_image], keep_prob: 1}) text = text_list[0].tolist() return text if __name__ == '__main__': train = 1 if train == 0: number = ['0','1','2','3','4','5','6','7','8','9'] #alphabet = ['a','b','c','d','e','f','g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v','w','x','y','z'] #ALPHABET = ['A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P','Q','R','S','T','U','V','W','X','Y','Z'] text, image = gen_captcha_text_and_image() print("验证码图像channel:", image.shape) # (60, 160, 3) # 图像大小 IMAGE_HEIGHT = 60 IMAGE_WIDTH = 160 MAX_CAPTCHA = len(text) print("验证码文本最长字符数", MAX_CAPTCHA) # 文本转向量 #char_set = number + alphabet + ALPHABET + ['_'] # 如果验证码长度小于4, '_'用来补齐 char_set = number CHAR_SET_LEN = len(char_set) X = tf.placeholder(tf.float32, [None, IMAGE_HEIGHT*IMAGE_WIDTH]) Y = tf.placeholder(tf.float32, [None, MAX_CAPTCHA*CHAR_SET_LEN]) keep_prob = tf.placeholder(tf.float32) # dropout train_crack_captcha_cnn() if train == 1: number = ['0','1','2','3','4','5','6','7','8','9'] IMAGE_HEIGHT = 60 IMAGE_WIDTH = 160 char_set = number CHAR_SET_LEN = len(char_set) text, image = gen_captcha_text_and_image() f = plt.figure() ax = f.add_subplot(111) ax.text(0.1, 0.9,text, ha='center', va='center', transform=ax.transAxes) plt.imshow(image) plt.show() MAX_CAPTCHA = len(text) image = convert2gray(image) image = image.flatten() / 255 X = tf.placeholder(tf.float32, [None, IMAGE_HEIGHT*IMAGE_WIDTH]) Y = tf.placeholder(tf.float32, [None, MAX_CAPTCHA*CHAR_SET_LEN]) keep_prob = tf.placeholder(tf.float32) # dropout predict_text = crack_captcha(image) print("正确: {} 预测: {}".format(text, predict_text))