基于tensorflow的验证码识别
基于tensorflow的验证码识别
背景介绍：
验证码图片样例： 
python库： tensorflow， opencv, pandas， gpu机器。
训练集： 10w 图片, 200step左右开始收敛。
策略： 切分图片，训练单字母识别。预测时也是同样切分。（ps:不切分训练及识别，跑了一夜，没有收敛）
准确率： 在区分大小写的情况下，单字母识别率98%， 整体识别率75%+。
训练集生成代码（大部分验证码都是插件生成，尽量找到生成方式，不然标注会很费力）：
package com;
import java.awt.Color;
import java.io.File;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.OutputStream;
import java.util.Random;
import org.patchca.color.ColorFactory;
import org.patchca.filter.predefined.CurvesRippleFilterFactory;
import org.patchca.filter.predefined.DiffuseRippleFilterFactory;
import org.patchca.filter.predefined.DoubleRippleFilterFactory;
import org.patchca.filter.predefined.MarbleRippleFilterFactory;
import org.patchca.filter.predefined.WobbleRippleFilterFactory;
import org.patchca.service.ConfigurableCaptchaService;
import org.patchca.utils.encoder.EncoderHelper;
import org.patchca.word.RandomWordFactory;
public class CreatePatcha {
private static Random random = new Random();
private static ConfigurableCaptchaService cs = new ConfigurableCaptchaService();
static {
// cs.setColorFactory(new SingleColorFactory(new Color(25, 60, 170)));
cs.setColorFactory(new ColorFactory() {
@Override
public Color getColor(int x) {
int[] c = new int[3];
int i = random.nextInt(c.length);
for (int fi = 0; fi < c.length; fi++) {
if (fi == i) {
c[fi] = random.nextInt(71);
} else {
c[fi] = random.nextInt(256);
}
}
return new Color(c[0], c[1], c[2]);
}
});
RandomWordFactory wf = new RandomWordFactory();
// wf.setCharacters("23456789abcdefghigklmnpqrstuvwxyzABCDEFGHIGKLMNPQRSTUVWXYZ");
wf.setCharacters("0123456789abcdefghigklmnopqrstuvwxyzABCDEFGHIGKLMNOPQRSTUVWXYZ");
wf.setMaxLength(4);
wf.setMinLength(4);
cs.setWordFactory(wf);
}
public static void main(String[] args) throws IOException {
for (int i = 0; i < 100; i++) {
switch (random.nextInt(5)) {
case 0:
cs.setFilterFactory(new CurvesRippleFilterFactory(cs
.getColorFactory()));
break;
case 1:
cs.setFilterFactory(new MarbleRippleFilterFactory());
break;
case 2:
cs.setFilterFactory(new DoubleRippleFilterFactory());
break;
case 3:
cs.setFilterFactory(new WobbleRippleFilterFactory());
break;
case 4:
cs.setFilterFactory(new DiffuseRippleFilterFactory());
break;
}
OutputStream out = new FileOutputStream(new File(i + ".png"));
String token = EncoderHelper.getChallangeAndWriteImage(cs, "png",
out);
out.close();
File f = new File(i+".png");
f.renameTo(new File("checkdata/" + token +"_" + i+".png"));
System.out.println(i+"验证码=" + token);
}
}
}
基于tf的神经网络训练代码（文件1读取训练集）：
#coding:utf-8 # from captcha.image import ImageCaptcha # pip install captcha import numpy as np import matplotlib.pyplot as plt from PIL import Image import random,time # 验证码中的字符, 就不用汉字了 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'] global total_index total_index = 0 global total_index_test total_index_test = 0 import os.path testDir = "testchars_padding" trainDir = "trainchars_padding" fileList = os.listdir(trainDir) testFileList = os.listdir(testDir) # 验证码一般都无视大小写；验证码长度4个字符 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 # 生成字符对应的验证码 def gen_captcha_text_and_image(train=True): global total_index global total_index_test if train: dir = trainDir captcha_text = fileList[total_index][5:6] captcha_image = Image.open(dir + "/" + fileList[total_index]).convert("RGB") captcha_image = np.array(captcha_image) total_index = (total_index + 1) % len(fileList) if(total_index % 10000 == 0): print('total_index:%d' % (total_index)) else: dir = testDir # print(total_index_test) captcha_text = testFileList[total_index_test][5:6] captcha_image = Image.open(dir + "/" + testFileList[total_index_test]).convert("RGB") captcha_image = np.array(captcha_image) total_index_test = (total_index_test + 1) % len(testFileList) return captcha_text, captcha_image
基于tf的神经网络训练代码（文件2，模型及训练）：
#coding:utf-8
from gen_captcha import gen_captcha_text_and_image
from gen_captcha import number
from gen_captcha import alphabet
from gen_captcha import ALPHABET
import numpy as np
import tensorflow as tf
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
text, image = gen_captcha_text_and_image()
print("验证码图像channel:", image.shape) # (70, 160, 3)
# 图像大小
IMAGE_HEIGHT = 70
IMAGE_WIDTH = 70
MAX_CAPTCHA = len(text)
print("验证码文本最长字符数", MAX_CAPTCHA) # 验证码最长4字符; 我全部固定为4,可以不固定. 如果验证码长度小于4，用'_'补齐
# 把彩色图像转为灰度图像（色彩对识别验证码没有什么用）
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
"""
cnn在图像大小是2的倍数时性能最高, 如果你用的图像大小不是2的倍数，可以在图像边缘补无用像素。
np.pad(image【,((2,3),(2,2)), 'constant', constant_values=(255,)) # 在图像上补2行，下补3行，左补2行，右补2行
"""
# 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 + alphabet + ALPHABET # 如果验证码长度小于4, '_'用来补齐
CHAR_SET_LEN = len(char_set) #26*2+10+1=63
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))
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, train = True):
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(70, 160, 3)
def wrap_gen_captcha_text_and_image(train):
while True:
text, image = gen_captcha_text_and_image(train)
if image.shape == (70, 70, 3):
return text, image
for i in range(batch_size):
text, image = wrap_gen_captcha_text_and_image(train)
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
####################################################################
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
# 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([9*9*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)
#out = tf.nn.softmax(out)
return out
# 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))
with tf.device('/gpu:0'):
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=output, labels=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()
config = tf.ConfigProto(allow_soft_placement=True)
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
sess.run(tf.global_variables_initializer())
step = 0
while True:
batch_x, batch_y = get_next_batch(256)
_, loss_ = sess.run([optimizer, loss], feed_dict={X: batch_x, Y: batch_y, keep_prob: 0.75})
# 每100 step计算一次准确率
if step % 100 == 0:
batch_x_test, batch_y_test = get_next_batch(100, False)
acc = sess.run(accuracy, feed_dict={X: batch_x_test, Y: batch_y_test, keep_prob: 1.})
print('step:%d,loss:%g' % (step, loss_))
print('step:%d,acc:%g'%(step, acc))
# 如果准确率大于50%,保存模型,完成训练
if acc > 0.98:
saver.save(sess, "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, tf.train.latest_checkpoint('.'))
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()
vector = np.zeros(MAX_CAPTCHA*CHAR_SET_LEN)
i = 0
for n in text:
vector[i*CHAR_SET_LEN + n] = 1
i += 1
return vec2text(vector)
if __name__ == '__main__':
#text, image = gen_captcha_text_and_image()
#image = convert2gray(image)
#image = image.flatten() / 255
#predict_text = crack_captcha(image)
#print("正确: {} 预测: {}".format(text, predict_text))
train_crack_captcha_cnn()
基于tf的神经网络训练代码（文件3，验证结果）：
#coding:utf-8 import numpy as np import matplotlib.pyplot as plt from PIL import Image import random,time import tensorflow as tf import os.path import cv2 from pandas import DataFrame from tensorflow_cnn import crack_captcha_cnn from tensorflow_cnn import X from tensorflow_cnn import Y from tensorflow_cnn import keep_prob from tensorflow_cnn import convert2gray os.environ["CUDA_VISIBLE_DEVICES"] = "3" 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'] # 图像大小 IMAGE_HEIGHT = 70 IMAGE_WIDTH = 70 MAX_CAPTCHA = 1 char_set = number + alphabet + ALPHABET # 如果验证码长度小于4, '_'用来补齐 CHAR_SET_LEN = len(char_set) #26*2+10+1=63 def splitImage(item): img = cv2.imread(item, 0) retval, img_black = cv2.threshold(img, 250, 255, cv2.THRESH_BINARY) img_black = cv2.bitwise_not(img_black) # cv2.imwrite('../test'+item[:-4]+'_black'+item[-4:],img_black) img_black = DataFrame(img_black) img_black.ix[70] = img_black.sum() / 255 numlist1 = [i + 25 for i in range(31)] numlist2 = [i + 65 for i in range(31)] numlist3 = [i + 105 for i in range(31)] def getpoint(ser, m): p = [] if m == 0: for i in range(len(ser)): if i in numlist1 and ser[i] <= m: p.append(i) if i in numlist2 and ser[i] <= m: p.append(i) if i in numlist3 and ser[i] <= m: p.append(i) else: for i in range(len(ser)): if i in numlist1 and ser[i - 1] <= m and ser[i] <= m and ser[i + 1] <= m: p.append(i) if i in numlist2 and ser[i - 1] <= m and ser[i] <= m and ser[i + 1] <= m: p.append(i) if i in numlist3 and ser[i - 1] <= m and ser[i] <= m and ser[i + 1] <= m: p.append(i) try: p1 = [] p2 = [] p3 = [] for i in p: if i <= 60: p1.append(i) if i > 60 and i <= 100: p2.append(i) if i > 100: p3.append(i) s1 = p1[int(len(p1) / 2)] s2 = p2[int(len(p2) / 2)] s3 = p3[int(len(p3) / 2)] except: s1, s2, s3 = 40, 80, 120 return [s1, s2, s3] s = getpoint(img_black.ix[70], 0) if s == None: s = getpoint(img_black.ix[70], 8) img = cv2.imread(item) # img1 = Image.fromarray((img[:, 0:s[0], :])) # img2 = Image.fromarray((img[:, s[0]:s[1], :])) # img3 = Image.fromarray((img[:, s[1]:s[2], :])) # img4 = Image.fromarray((img[:, s[2]:, :])) img1 = (img[:, 0:s[0], :]) img2 = (img[:, s[0]:s[1], :]) img3 = (img[:, s[1]:s[2], :]) img4 = (img[:, s[2]:, :]) return [padding(img1), padding(img2), padding(img3), padding(img4)] def get_imlist(path): return [os.path.join(path, f) for f in os.listdir(path) if f.endswith('.png')] def padding(img): # img = cv2.imread(item) base = np.zeros(4900 * 3).reshape((IMAGE_HEIGHT, IMAGE_WIDTH, 3)) base += 255 m = img.shape[1] start = int((70 - m) / 2) end = start + m base[:, start:end, :] = img return base # 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)) return "".join(text) if __name__ == '__main__': output = crack_captcha_cnn() saver = tf.train.Saver() with tf.Session() as sess: saver.restore(sess, tf.train.latest_checkpoint('.')) imglist = get_imlist('./data/') total = 0 right = 0 for item in imglist: print("*************************************: {}".format(item)) imgs = splitImage(item) str = "" for img in imgs: # image = Image.fromarray(img) image = convert2gray(img) image = image.flatten() / 255 predict = tf.argmax(tf.reshape(output, [-1, MAX_CAPTCHA, CHAR_SET_LEN]), 2) text_list = sess.run(predict, feed_dict={X: [image], keep_prob: 1}) text = text_list[0].tolist() vector = np.zeros(MAX_CAPTCHA * CHAR_SET_LEN) i = 0 for n in text: vector[i * CHAR_SET_LEN + n] = 1 i += 1 predict_text = vec2text(vector) str = str + predict_text total = total + 1 if(str in item): right = right + 1 print("正确: {} 预测: {} 结果: {} 正确: {} 总数: {}".format(item, str, str in item, right, total))
结果：
参考：https://zhuanlan.zhihu.com/p/25779608?group_id=825335754321457152
