python入门教程之识别验证码

那我来讲解关于“Python入门教程之识别验证码”的攻略。

1. 前言

验证码是目前防止自动化机器人攻击的一种重要方式。而在自动化测试、爬虫等场景下，我们又需要自动识别验证码。因此，学习如何识别验证码也是学习Python的重要一环。

2. 主要技术

本教程将采用Python 3.7版本，涉及到如下技术：

• 图像处理

• 机器学习

• 神经网络

3. 环境和库的准备

首先我们需要准备开发环境，这里我们建议使用Anaconda来安装Python。然后我们需要安装相关库来进行图像处理、机器学习以及神经网络的开发。这里数Python基本的库：NumPy、SciPy等。在机器学习和神经网络的开发中，我们推荐使用tensorflow。

4.验证码的识别过程

验证码的识别主要涉及以下几个步骤：

• 数据集的准备

• 图像预处理

• 训练模型

• 验证码测试

4.1 数据集的准备

为了训练模型,我们需要一个包含验证码的数据集，这里我们可以直接通过网络爬虫爬取，或者通过收集公开数据集来获取。可以通过以下方式获取训练集：

import requests
import os

def save_image(url, filename):
    with open(filename, "wb") as f:
        f.write(requests.get(url).content)

def download_captcha(count):
    for i in range(count):
        url = "http://www.example.com/captcha/{0}.png".format(str(i))
        filename = os.path.join("captcha", "{0}.png".format(str(i)))
        save_image(url, filename)

if __name__ == '__main__':
    download_captcha(10000)   #下载10000个验证码

4.2 图像预处理

预处理是为了让原始图像更适合机器学习或神经网络算法的处理，预处理中可以包括以下技术：

• 二值化

• 去噪

• 分割

import glob
import cv2

def preprocess_image(image_path):
    gray = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
    gray = cv2.resize(gray, (200, 60))
    _, binary = cv2.threshold(gray, 180, 255, cv2.THRESH_BINARY)
    kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
    binary = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel)
    binary = cv2.medianBlur(binary, 3)
    return binary

if __name__ == '__main__':
    paths = glob.glob(os.path.join("captcha", "*.png"))
    for image_path in paths:
        binary = preprocess_image(image_path)
        cv2.imwrite(os.path.join("captcha_preprocessed", os.path.basename(image_path)), binary)

4.3 训练模型

这里我们使用tensorflow来训练一个卷积神经网络模型（CNN），用于识别验证码：

import tensorflow as tf
import os

def train_cnn():
    char_set = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
    char_set_len = len(char_set)
    input_height, input_width = 60, 200
    input_channel = 1
    batch_size = 100
    epochs = 100
    output_node_names = "output"
    model_dir = os.path.join("captcha_model", "model.ckpt")

    X = tf.placeholder(tf.float32, [None, input_height, input_width, input_channel])
    Y = tf.placeholder(tf.float32, [None, char_set_len * 4])

    logits = cnn(X, char_set_len)
    loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=logits, labels=Y))
    optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)

    predict = tf.reshape(logits, [-1, 4, char_set_len])
    max_pred = tf.argmax(predict, axis=2)
    decoded_pred = tf.reduce_join(tf.transpose(tf.argmax(predict, axis=2), perm=[1, 0]), 1)

    saver = tf.train.Saver()
    with tf.Session() as session:
        session.run(tf.global_variables_initializer())
        for epoch in range(epochs):
            avg_loss = 0.
            total_batch = int(len(X_train) / batch_size)
            for i in range(total_batch):
                batch_X = X_train[i*batch_size:(i+1)*batch_size]
                batch_Y = Y_train[i*batch_size:(i+1)*batch_size]
                _, c = session.run([optimizer, loss], feed_dict={X: batch_X, Y: batch_Y})
                avg_loss += c / total_batch
            print("Epoch:", '%d' % (epoch + 1), "cost=", "{:.9f}".format(avg_loss))

        saver.save(session, model_dir)
        print("Model saved!")

if __name__ == '__main__':
    train_cnn()

4.4 验证码测试

训练好模型后，我们可以用训练好的模型去测试验证码的识别效果了。验证码测试主要步骤如下：

• 从验证码中分割出4个字符

• 对每个字符进行预处理

• 利用模型进行识别

import glob
import cv2
import os
import numpy as np
import tensorflow as tf

def test_captcha():
    char_set = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
    char_set_len = len(char_set)
    input_height, input_width = 60, 200
    input_channel = 1
    model_dir = os.path.join("captcha_model", "model.ckpt")

    X = tf.placeholder(tf.float32, [None, input_height, input_width, input_channel])
    logits = cnn(X, char_set_len)
    predict = tf.reshape(logits, [-1, 4, char_set_len])
    max_pred = tf.argmax(predict, axis=2)
    decoded_pred = tf.reduce_join(tf.transpose(tf.argmax(predict, axis=2), perm=[1, 0]), 1)

    saver = tf.train.Saver()
    with tf.Session() as session:
        saver.restore(session, model_dir)
        paths = glob.glob(os.path.join("captcha", "*.png"))
        for image_path in paths:
            image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
            _, binary = cv2.threshold(image, 100, 255, cv2.THRESH_BINARY)
            contours, hierarchy = cv2.findContours(binary, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
            c_len = len(contours)
            if c_len != 4:
                print("can't recognize captcha.")
                continue
            part_digits = []
            for j in range(c_len):
                x, y, w, h = cv2.boundingRect(contours[j])
                part_image = binary[y:y+h, x:x+w]
                part_image = cv2.resize(part_image, (input_width, input_height))
                part_digits.append(part_image)
            part_digits_array = np.array(part_digits) / 255.0
            predict_res = session.run(decoded_pred, feed_dict={X: part_digits_array[:, :, :, np.newaxis]})
            print(predict_res)

if __name__ == '__main__':
    test_captcha()

以上就是关于“Python入门教程之识别验证码”的攻略了。