毕设成品 基于机器视觉的验证码识别研究与实现

---

0 项目说明

基于机器视觉的验证码识别研究与实现

提示：适合用于课程设计或毕业设计，工作量达标，源码开放

---

1 使用方法

1. 爬取验证码
2. 对图像做处理并切分
3. 手工标注数据
4. 导入训练集
5. 使用测试集

2 前期准备

1.Image (图像处理库)

• jpeg
• zlib
• PIL

2.numpy (数学处理库)
3.ImageEnhance (图像处理库)

enhancer = ImageEnhance.Contrast(img)    # 增加对比对
img = enhancer.enhance(2)
enhancer = ImageEnhance.Sharpness(img)   # 锐化
img = enhancer.enhance(2)
enhancer = ImageEnhance.Brightness(img)  # 增加亮度
img = enhancer.enhance(2)

3 图像处理

静态图片

1. 清除图片噪点
2. 清除图片干扰线
3. 切割图片
4. 信息输出

动态图片

1. 按帧转存 GIF
2. 读取每个 GIF 的 Duration 属性
3. 找到 Duration 最长的图片，后同静态图片处理

4 识别算法

KNN

# kNN algorithm
def classify0(inX, dataSet, labels, k):
    dataSetSize = dataSet.shape[0]
    diffMat = tile(inX, (dataSetSize, 1)) - dataSet
    sqDiffMat = diffMat ** 2
    sqDistances = sqDiffMat.sum(axis=1)
    distances = sqDistances ** 0.5
    sortedDistIndicies = distances.argsort()
    classCount = {}
    for i in range(k):
        voteIlabel = labels[sortedDistIndicies[i]]  # changed
        classCount[voteIlabel] = classCount.get(voteIlabel, 0) + 1
    sortedClassCount = sorted(classCount.iteritems(), key=operator.itemgetter(1), reverse=True)
    return sortedClassCount[0][0]

SVM
根据算法的性质，可以问题设定成一个二分类问题：识别数字1和2（当然也可以是其他的任意两个数字)。

5 项目源码

# __author__ = 'Yaoshi'
# -*- coding: utf-8 -*-
# This py file is to the split the captcha into single alphabet or number
from __future__ import division
import time
import urllib2
import socks
from sockshandler import SocksiPyHandler
from PIL import Image, ImageEnhance, ImageFilter, ImageGrab


class Decaptcha:
    def __init__(self, new_img_id, counter, number):
        while counter < number:
            print 'Now processing pic no %d' % counter
            tmp_file_name = self.crawler(counter)
            temp_img_id = self.img2binary(img=tmp_file_name, new_img_id=new_img_id, counter=counter)
            print 'Get %d alphabet or number', temp_img_id - new_img_id
            new_img_id = temp_img_id
            counter += 1
            time.sleep(1)
        print 'Cut out success ratio:', new_img_id / (counter * 4)

    def crawler(self, counter):
        ip = ''
        port = 1080
        url = ''
        filename = "images/MJ%d.jpg" % counter
        opener = urllib2.build_opener(SocksiPyHandler(socks.SOCKS5, ip, port))
        opener.addheaders = [('User-Agent',
                              'Mozilla/5.0 (Windows NT 6.1; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/34.0.1847.137 Safari/537.36')]
        response = opener.open(url)
        htmlData = response.read()
        f = open(filename, 'w')
        f.write(htmlData)
        f.close()
        return filename

    def add_background(self, image, counter):
        img = Image.open(image)
        x, y = img.size
        p = Image.new('RGBA', img.size, (255, 255, 255))
        p.paste(img, (0, 0, x, y), img)
        p.save("bk-images/MJbackground-%d.png" % counter, "PNG")
        return p

    def img2binary(self, img, new_img_id, counter):
        img = self.add_background(image=img, counter=counter)
        assert isinstance(img, Image.Image)
        img = img.filter(ImageFilter.MedianFilter(3))  # use median filter to de-noise
        img.save("filter-images/%d.png" % counter, "PNG")
        enhancer = ImageEnhance.Contrast(img)
        img = enhancer.enhance(2)
        enhancer = ImageEnhance.Sharpness(img)
        img = enhancer.enhance(2)
        enhancer = ImageEnhance.Brightness(img)
        img = enhancer.enhance(2)
        img = img.convert("L")  # transfer the image into gray-scale
        img.save("gray-images/%d.png" % counter, "PNG")
        length = img.size[0]
        width = img.size[1]     # find the length and the width of the image
        # print 'size-length:', length, 'width', width
        counter = 0
        num_of_valid_pix = []   # this data structure is to store the number of valid pixels for each column.
        pixdata = img.load()    # load the image data into the @pixdata
        # retrival all the pixels in the image
        for x in range(0, length):
            for y in range(0, width):
                # change the valve
                if pixdata[x, y] < 200:
                    counter += 1
                    pixdata[x, y] = 0    # reset the pixdata to binary form, 1 represents for valid pixel
                else:
                    pixdata[x, y] = 255  # reset the pixdata to binary form, 0 represents for invalid pixel
            num_of_valid_pix.append(counter)
            counter = 0  # this counter is used to count the number of the pixel for each row
        letter_col_id = []
        i = 0
        # the following part is to separate the letters out from the given CAPTCHA.
        while i in range(len(num_of_valid_pix)):
            letter_id = []  # @letter_id stores the cols for each letter
            # letter feature: there must be blank cols that contains no valid pixels in the column
            while num_of_valid_pix[i] != 0:
                letter_id.append(i)
                i += 1
            if letter_id:
                letter_col_id.append(letter_id)
            i += 1
        # check the num of lines for each letter
        numofLetters = len(letter_col_id)
        # this part is dealing with the saparated
        for j in range(numofLetters):
            colsForLetter = len(letter_col_id[j])
            # if colsForLetter in range(3, 25):
            if colsForLetter in range(5, 14):
                file = open("trainingdigit/demo%d.txt" % new_img_id, 'w')
                # listbuffer = []
                newimg = Image.new("L", (len(letter_col_id[j]), width))
                # newimg = newimg.load()
                for y in range(width):
                    # rowbuffer = []
                    i = 0
                    for x in letter_col_id[j]:
                        # rowbuffer.append(pixdata[x, y])
                        # newimg[i, y] = pixdata[x, y]
                        if pixdata[x, y] == 255:
                            file.write("0")  # 0 for there's a invalid digit
                        elif pixdata[x, y] == 0:
                            file.write("1")  # 1 for there's a valid digit
                        newimg.putpixel([i, y], pixdata[x, y])
                        i += 1
                    # listbuffer.append(rowbuffer)
                    file.write("\n")
                file.close()
                newimg.save("trainingdigit/letter_%d.png" % new_img_id, "PNG")
                new_img_id += 1
        return new_img_id


if __name__ == '__main__':
    handle_class = Decaptcha(new_img_id=0, counter=0, number=400)

6 最后

**项目分享: ** https://gitee.com/asoonis/htw