TensorFlow實現驗證碼識別（十七）
阿新 • • 發佈：2019-02-14
TensorFlow實現驗證碼的識別
使用的生成驗證碼庫是captcha。
首先是驗證碼的生成，先生成一個驗證碼。
程式碼：
import random import tensorflow as tf from captcha.image import ImageCaptcha import matplotlib.pyplot as plt import PIL.Image import numpy as np 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_alphabet(chat_set = number , chat_size = 4): chat_test = [] for i in range(chat_size): chat = random.choice(chat_set) chat_test.append(chat) returnchat_test def get_text_image(): image = ImageCaptcha() text = random_alphabet() for i in range(len(text)): text[i] = str(text[i]) new_text = ''.join(text) text = new_text captcha = image.generate(text) captcha_image = PIL.Image.open(captcha) captcha_image = np.array(captcha_image) returntext ,captcha_image def show(): text , image = get_text_image() f = plt.figure() ax = f.add_subplot(111) plt.imshow(image) plt.show()
這幾個就是生成驗證碼的函式，放在image_Create.py檔案下。
解釋下程式碼吧
首先是先得到要一個驗證碼上的數字：
因為這個如果識別的數字或字母特別多，要訓練很久，於是就只是進行數字的驗證碼識別。
def random_alphabet(chat_set = number , chat_size = 4): chat_test = [] for i in range(chat_size): chat = random.choice(chat_set) chat_test.append(chat) return chat_test
random.choice就是在set集合中選擇4個驗證碼的數字加入到數字集。返回即可。
def get_text_image(): image = ImageCaptcha() text = random_alphabet() for i in range(len(text)): text[i] = str(text[i]) new_text = ''.join(text) text = new_text captcha = image.generate(text) captcha_image = PIL.Image.open(captcha) captcha_image = np.array(captcha_image) return text ,captcha_image
這個函式就是返回一個驗證碼數字和一個圖片，驗證碼的圖片。
captcha = image.generate(text)這個就是生成驗證碼的圖片了。後面就是一些固定操作了。
def show(): text , image = get_text_image() f = plt.figure() ax = f.add_subplot(111) plt.imshow(image) plt.show()
這個就是一個現實圖片的函式。加一個迴圈把list裡面的數字都變成一個字串，然後把整個list轉換成一個string，最後用imshow把圖片放進去，show顯示。
接下來就是進行一個訓練了。
import tensorflow as tf import numpy as np import image_Create import prev import matplotlib.pyplot as plt def train_cell(x , w_alpha = 0.01 , b_alpha = 0.1): x = tf.reshape(x , shape=[-1 , image_height , image_width , 1]) 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_cell) 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_cell) 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_cell) 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_cell) w_out = tf.Variable(w_alpha*tf.random_normal([1024, Max_text * Char_size])) b_out = tf.Variable(b_alpha*tf.random_normal([Max_text * Char_size])) out = tf.add(tf.matmul(dense, w_out), b_out) return out pass def train_CNN(x , y): output = train_cell(x) loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=output , labels=y)) optimer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss) predict = tf.reshape(output, [-1, Max_text, Char_size]) max_idx_p = tf.argmax(predict, 2) max_idx_l = tf.argmax(tf.reshape(y, [-1, Max_text , Char_size]), 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([optimer, loss], feed_dict={x: batch_x, y: batch_y, keep_cell: 0.75}) print('第',step,'步','損失', loss_) # 每100 step計算一次準確率 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_cell: 1.}) print('第',step,'步', '準確',acc) if step > 1000: saver.save(sess, "./model/crack_capcha.model", global_step=step) break step+=1 pass def get_next_batch(batch_size = 128): batch_x = np.zeros([batch_size , image_height * image_width]) batch_y = np.zeros([batch_size , Max_text * Char_size]) for i in range(batch_size): text , image = image_Create.get_text_image() image = prev.convert_to_gray(image) batch_x[i ,:] = image.flatten()/225 batch_y[i ,:] = prev.text2vec(text,Max_text , Char_size) pass return batch_x , batch_y pass def crack_captcha(captcha_image): output = train_cell(x) saver = tf.train.Saver() with tf.Session() as sess: saver.restore(sess, "./model/crack_capcha.model-1010") predict = tf.argmax(tf.reshape(output, [-1, Max_text, Char_size]), 2) text_list = sess.run(predict, feed_dict={x: [captcha_image], keep_cell: 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 = image_Create.get_text_image() image_height , image_width = (60,160) char_set = number Max_text = len(text) Char_size = len(char_set) x = tf.placeholder(tf.float32 , [None , image_height*image_width]) y = tf.placeholder(tf.float32 , [None , Max_text*Char_size]) keep_cell = tf.placeholder(tf.float32 ) train_CNN(x , y) if train == 1: number = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9'] image_height = 60 image_width = 160 char_set = number Char_size = len(char_set) text, image = image_Create.get_text_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_text = len(text) image = prev.convert_to_gray(image) image = image.flatten() / 255 x = tf.placeholder(tf.float32, [None, image_height * image_width]) y = tf.placeholder(tf.float32, [None, Max_text * Char_size]) keep_cell = tf.placeholder(tf.float32) # dropout predict_text = crack_captcha(image) print("正確: {} 預測: {}".format(text, predict_text))
首先要定義網路。
def train_cell(x , w_alpha = 0.01 , b_alpha = 0.1): x = tf.reshape(x , shape=[-1 , image_height , image_width , 1]) 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_cell) 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_cell) 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_cell) 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_cell) w_out = tf.Variable(w_alpha*tf.random_normal([1024, Max_text * Char_size])) b_out = tf.Variable(b_alpha*tf.random_normal([Max_text * Char_size])) out = tf.add(tf.matmul(dense, w_out), b_out) return out pass
之前都講過了，應該都有b數的。三層卷積網路，兩層全連線網路。
之後就是一個損失函數了：
def train_CNN(x , y): output = train_cell(x) loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=output , labels=y)) optimer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss) predict = tf.reshape(output, [-1, Max_text, Char_size]) max_idx_p = tf.argmax(predict, 2) max_idx_l = tf.argmax(tf.reshape(y, [-1, Max_text , Char_size]), 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([optimer, loss], feed_dict={x: batch_x, y: batch_y, keep_cell: 0.75}) print('第',step,'步','損失', loss_) # 每100 step計算一次準確率 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_cell: 1.}) print('第',step,'步', '準確',acc) # 如果準確率大於50%,儲存模型,完成訓練 if step > 1000: saver.save(sess, "./model/crack_capcha.model", global_step=step) break step+=1 pass
可以看到使用的優化器並不是梯度下降，但是使用梯度下降也可以。。。但是要把學習率設定的很小，之前都試過，這個是最快的了。儲存模型。。。應該懂的。
這裡要用到幾個圖片的預處理：
import numpy as np import tensorflow as tf def convert_to_gray(image): if len(image.shape) > 2: gray = np.mean(image , -1) return gray else: return image def text2vec(text , max_text , char_size): text_len = len(text) vector = np.zeros(max_text*char_size) for i, c in enumerate(text): idx = i * char_size + int(c) vector[idx] = 1 return vector
np.mean(image,-1)是為了轉換成灰度圖
會放在pev.py下
def get_next_batch(batch_size = 128): batch_x = np.zeros([batch_size , image_height * image_width]) batch_y = np.zeros([batch_size , Max_text * Char_size]) for i in range(batch_size): text , image = image_Create.get_text_image() image = prev.convert_to_gray(image) batch_x[i ,:] = image.flatten()/225 batch_y[i ,:] = prev.text2vec(text,Max_text , Char_size) pass return batch_x , batch_y pass
這個是讀取模型。1010就是訓練了1010次後得到的模型。
def crack_captcha(captcha_image): output = train_cell(x) saver = tf.train.Saver() with tf.Session() as sess: saver.restore(sess, "./model/crack_capcha.model-1010") predict = tf.argmax(tf.reshape(output, [-1, Max_text, Char_size]), 2) text_list = sess.run(predict, feed_dict={x: [captcha_image], keep_cell: 1}) text = text_list[0].tolist() return text
接下來就是運行了：
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 = image_Create.get_text_image() image_height , image_width = (60,160) char_set = number Max_text = len(text) Char_size = len(char_set) x = tf.placeholder(tf.float32 , [None , image_height*image_width]) y = tf.placeholder(tf.float32 , [None , Max_text*Char_size]) keep_cell = tf.placeholder(tf.float32 ) train_CNN(x , y) if train == 1: number = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9'] image_height = 60 image_width = 160 char_set = number Char_size = len(char_set) text, image = image_Create.get_text_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_text = len(text) image = prev.convert_to_gray(image) image = image.flatten() / 255 x = tf.placeholder(tf.float32, [None, image_height * image_width]) y = tf.placeholder(tf.float32, [None, Max_text * Char_size]) keep_cell = tf.placeholder(tf.float32) # dropout predict_text = crack_captcha(image) print("正確: {} 預測: {}".format(text, predict_text))
一開始訓練很慢的，訓練1010次後得到的模型的準確率是0.72。
還算過得去吧。。。。。
def train_cell(x , w_alpha = 0.01 , b_alpha = 0.1): x = tf.reshape(x , shape=[-1 , image_height , image_width , 1]) 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_cell) 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_cell) 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_cell) 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_cell) w_out = tf.Variable(w_alpha*tf.random_normal([1024, Max_text * Char_size])) b_out = tf.Variable(b_alpha*tf.random_normal([Max_text * Char_size])) out = tf.add(tf.matmul(dense, w_out), b_out) return out pass
def train_CNN(x , y): output = train_cell(x) loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=output , labels=y)) optimer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss) predict = tf.reshape(output, [-1, Max_text, Char_size]) max_idx_p = tf.argmax(predict, 2) max_idx_l = tf.argmax(tf.reshape(y, [-1, Max_text , Char_size]), 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([optimer, loss], feed_dict={x: batch_x, y: batch_y, keep_cell: 0.75}) print('第',step,'步','損失', loss_) # 每100 step計算一次準確率 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_cell: 1.}) print('第',step,'步', '準確',acc) # 如果準確率大於50%,儲存模型,完成訓練 if step > 1000: saver.save(sess, "./model/crack_capcha.model", global_step=step) break step+=1 pass