您当前的位置:首页 > 计算机 > 编程开发 > Python

Python——破解极验滑动验证码

时间:10-29来源:作者:点击数:
CDSY,CDSY.XYZ

极验滑动验证码

以上图片是最典型的要属于极验滑动认证了,极验官网:http://www.geetest.com/。

现在极验验证码已经更新到了 3.0 版本,截至 2017 年 7 月全球已有十六万家企业正在使用极验,每天服务响应超过四亿次,广泛应用于直播视频、金融服务、电子商务、游戏娱乐、政府企业等各大类型网站

对于这类验证,如果我们直接模拟表单请求,繁琐的认证参数与认证流程会让你蛋碎一地,我们可以用selenium驱动浏览器来解决这个问题,大致分为以下几个步骤

1、输入用户名,密码

2、点击按钮验证,弹出没有缺口的图

3、获得没有缺口的图片

4、点击滑动按钮,弹出有缺口的图

5、获得有缺口的图片

6、对比两张图片,找出缺口,即滑动的位移

7、按照人的行为行为习惯,把总位移切成一段段小的位移

8、按照位移移动

9、完成登录

实现

位移移动需要的基础知识

位移移动相当于匀变速直线运动,类似于小汽车从起点开始运行到终点的过程(首先为匀加速,然后再匀减速)。

其中a为加速度,且为恒量(即单位时间内的加速度是不变的),t为时间

位移移动的代码实现

def get_track(distance):
    '''
    拿到移动轨迹,模仿人的滑动行为,先匀加速后匀减速
    匀变速运动基本公式:
    ①v=v0+at
    ②s=v0t+(1/2)at²
    ③v²-v0²=2as

    :param distance: 需要移动的距离
    :return: 存放每0.2秒移动的距离
    '''
    # 初速度
    v=0
    # 单位时间为0.2s来统计轨迹,轨迹即0.2内的位移
    t=0.1
    # 位移/轨迹列表,列表内的一个元素代表0.2s的位移
    tracks=[]
    # 当前的位移
    current=0
    # 到达mid值开始减速
    mid=distance * 4/5

    distance += 10  # 先滑过一点,最后再反着滑动回来

    while current < distance:
        if current < mid:
            # 加速度越小,单位时间的位移越小,模拟的轨迹就越多越详细
            a = 2  # 加速运动
        else:
            a = -3 # 减速运动

        # 初速度
        v0 = v
        # 0.2秒时间内的位移
        s = v0*t+0.5*a*(t**2)
        # 当前的位置
        current += s
        # 添加到轨迹列表
        tracks.append(round(s))

        # 速度已经达到v,该速度作为下次的初速度
        v= v0+a*t

    # 反着滑动到大概准确位置
    for i in range(3):
       tracks.append(-2)
    for i in range(4):
       tracks.append(-1)
    return tracks
对比两张图片,找出缺口
def get_distance(image1,image2):
    '''
      拿到滑动验证码需要移动的距离
      :param image1:没有缺口的图片对象
      :param image2:带缺口的图片对象
      :return:需要移动的距离
      '''
    # print('size', image1.size)

    threshold = 50
    for i in range(0,image1.size[0]):  # 260
        for j in range(0,image1.size[1]):  # 160
            pixel1 = image1.getpixel((i,j))
            pixel2 = image2.getpixel((i,j))
            res_R = abs(pixel1[0]-pixel2[0]) # 计算RGB差
            res_G = abs(pixel1[1] - pixel2[1])  # 计算RGB差
            res_B = abs(pixel1[2] - pixel2[2])  # 计算RGB差
            if res_R > threshold and res_G > threshold and res_B > threshold:
                return i  # 需要移动的距离
获得图片
def merge_image(image_file,location_list):
    """
     拼接图片
    :param image_file:
    :param location_list:
    :return:
    """
    im = Image.open(image_file)
    im.save('code.jpg')
    new_im = Image.new('RGB',(260,116))
    # 把无序的图片 切成52张小图片
    im_list_upper = []
    im_list_down = []
    # print(location_list)
    for location in location_list:
        # print(location['y'])
        if location['y'] == -58: # 上半边
            im_list_upper.append(im.crop((abs(location['x']),58,abs(location['x'])+10,116)))
        if location['y'] == 0:  # 下半边
            im_list_down.append(im.crop((abs(location['x']),0,abs(location['x'])+10,58)))

    x_offset = 0
    for im in im_list_upper:
        new_im.paste(im,(x_offset,0))  # 把小图片放到 新的空白图片上
        x_offset += im.size[0]

    x_offset = 0
    for im in im_list_down:
        new_im.paste(im,(x_offset,58))
        x_offset += im.size[0]
    new_im.show()
    return new_im

def get_image(driver,div_path):
    '''
    下载无序的图片  然后进行拼接 获得完整的图片
    :param driver:
    :param div_path:
    :return:
    '''
    time.sleep(2)
    background_images = driver.find_elements_by_xpath(div_path)
    location_list = []
    for background_image in background_images:
        location = {}
        result = re.findall('background-image: url\("(.*?)"\); background-position: (.*?)px (.*?)px;',background_image.get_attribute('style'))
        # print(result)
        location['x'] = int(result[0][1])
        location['y'] = int(result[0][2])

        image_url = result[0][0]
        location_list.append(location)

    print('==================================')
    image_url = image_url.replace('webp','jpg')
    # '替换url http://static.geetest.com/pictures/gt/579066de6/579066de6.webp'
    image_result = requests.get(image_url).content
    # with open('1.jpg','wb') as f:
    #     f.write(image_result)
    image_file = BytesIO(image_result) # 是一张无序的图片
    image = merge_image(image_file,location_list)

    return image
按照位移移动
 print('第一步,点击滑动按钮')
    ActionChains(driver).click_and_hold(on_element=element).perform()  # 点击鼠标左键,按住不放
    time.sleep(1)
    print('第二步,拖动元素')
    for track in track_list:
         ActionChains(driver).move_by_offset(xoffset=track, yoffset=0).perform() # 鼠标移动到距离当前位置(x,y)
    if l<100:
        ActionChains(driver).move_by_offset(xoffset=-2, yoffset=0).perform()
    else:
        ActionChains(driver).move_by_offset(xoffset=-5, yoffset=0).perform()
    time.sleep(1)
    print('第三步,释放鼠标')
    ActionChains(driver).release(on_element=element).perform()
详细代码
from selenium import webdriver
from selenium.webdriver.support.ui import WebDriverWait # 等待元素加载的
from selenium.webdriver.common.action_chains import ActionChains  #拖拽
from selenium.webdriver.support import expected_conditions as EC
from selenium.common.exceptions import TimeoutException, NoSuchElementException
from selenium.webdriver.common.by import By
from PIL import Image
import requests
import time
import re
import random
from io import BytesIO


def merge_image(image_file,location_list):
    """
     拼接图片
    :param image_file:
    :param location_list:
    :return:
    """
    im = Image.open(image_file)
    im.save('code.jpg')
    new_im = Image.new('RGB',(260,116))
    # 把无序的图片 切成52张小图片
    im_list_upper = []
    im_list_down = []
    # print(location_list)
    for location in location_list:
        # print(location['y'])
        if location['y'] == -58: # 上半边
            im_list_upper.append(im.crop((abs(location['x']),58,abs(location['x'])+10,116)))
        if location['y'] == 0:  # 下半边
            im_list_down.append(im.crop((abs(location['x']),0,abs(location['x'])+10,58)))

    x_offset = 0
    for im in im_list_upper:
        new_im.paste(im,(x_offset,0))  # 把小图片放到 新的空白图片上
        x_offset += im.size[0]

    x_offset = 0
    for im in im_list_down:
        new_im.paste(im,(x_offset,58))
        x_offset += im.size[0]
    new_im.show()
    return new_im

def get_image(driver,div_path):
    '''
    下载无序的图片  然后进行拼接 获得完整的图片
    :param driver:
    :param div_path:
    :return:
    '''
    time.sleep(2)
    background_images = driver.find_elements_by_xpath(div_path)
    location_list = []
    for background_image in background_images:
        location = {}
        result = re.findall('background-image: url\("(.*?)"\); background-position: (.*?)px (.*?)px;',background_image.get_attribute('style'))
        # print(result)
        location['x'] = int(result[0][1])
        location['y'] = int(result[0][2])

        image_url = result[0][0]
        location_list.append(location)

    print('==================================')
    image_url = image_url.replace('webp','jpg')
    # '替换url http://static.geetest.com/pictures/gt/579066de6/579066de6.webp'
    image_result = requests.get(image_url).content
    # with open('1.jpg','wb') as f:
    #     f.write(image_result)
    image_file = BytesIO(image_result) # 是一张无序的图片
    image = merge_image(image_file,location_list)

    return image

def get_track(distance):
    '''
    拿到移动轨迹,模仿人的滑动行为,先匀加速后匀减速
    匀变速运动基本公式:
    ①v=v0+at
    ②s=v0t+(1/2)at²
    ③v²-v0²=2as

    :param distance: 需要移动的距离
    :return: 存放每0.2秒移动的距离
    '''
    # 初速度
    v=0
    # 单位时间为0.2s来统计轨迹,轨迹即0.2内的位移
    t=0.2
    # 位移/轨迹列表,列表内的一个元素代表0.2s的位移
    tracks=[]
    # 当前的位移
    current=0
    # 到达mid值开始减速
    mid=distance * 7/8

    distance += 10  # 先滑过一点,最后再反着滑动回来
    # a = random.randint(1,3)
    while current < distance:
        if current < mid:
            # 加速度越小,单位时间的位移越小,模拟的轨迹就越多越详细
            a = random.randint(2,4)  # 加速运动
        else:
            a = -random.randint(3,5) # 减速运动

        # 初速度
        v0 = v
        # 0.2秒时间内的位移
        s = v0*t+0.5*a*(t**2)
        # 当前的位置
        current += s
        # 添加到轨迹列表
        tracks.append(round(s))

        # 速度已经达到v,该速度作为下次的初速度
        v= v0+a*t

    # 反着滑动到大概准确位置
    for i in range(4):
       tracks.append(-random.randint(2,3))
    for i in range(4):
       tracks.append(-random.randint(1,3))
    return tracks


def get_distance(image1,image2):
    '''
      拿到滑动验证码需要移动的距离
      :param image1:没有缺口的图片对象
      :param image2:带缺口的图片对象
      :return:需要移动的距离
      '''
    # print('size', image1.size)

    threshold = 50
    for i in range(0,image1.size[0]):  # 260
        for j in range(0,image1.size[1]):  # 160
            pixel1 = image1.getpixel((i,j))
            pixel2 = image2.getpixel((i,j))
            res_R = abs(pixel1[0]-pixel2[0]) # 计算RGB差
            res_G = abs(pixel1[1] - pixel2[1])  # 计算RGB差
            res_B = abs(pixel1[2] - pixel2[2])  # 计算RGB差
            if res_R > threshold and res_G > threshold and res_B > threshold:
                return i  # 需要移动的距离



def main_check_code(driver, element):
    """
     拖动识别验证码
    :param driver: 
    :param element: 
    :return: 
    """
    image1 = get_image(driver, '//div[@class="gt_cut_bg gt_show"]/div')
    image2 = get_image(driver, '//div[@class="gt_cut_fullbg gt_show"]/div')
    # 图片上 缺口的位置的x坐标

    # 2 对比两张图片的所有RBG像素点,得到不一样像素点的x值,即要移动的距离
    l = get_distance(image1, image2)
    print('l=',l)
    # 3 获得移动轨迹
    track_list = get_track(l)
    print('第一步,点击滑动按钮')
    ActionChains(driver).click_and_hold(on_element=element).perform()  # 点击鼠标左键,按住不放
    time.sleep(1)
    print('第二步,拖动元素')
    for track in track_list:
         ActionChains(driver).move_by_offset(xoffset=track, yoffset=0).perform()  # 鼠标移动到距离当前位置(x,y)
     time.sleep(0.002)
    # if l>100:

    ActionChains(driver).move_by_offset(xoffset=-random.randint(2,5), yoffset=0).perform()
    time.sleep(1)
    print('第三步,释放鼠标')
    ActionChains(driver).release(on_element=element).perform()
    time.sleep(5)


def main_check_slider(driver):
    """
    检查滑动按钮是否加载
    :param driver: 
    :return: 
    """
    while True:
        try :
            driver.get('http://www.cnbaowen.net/api/geetest/')
            element = WebDriverWait(driver, 30, 0.5).until(EC.element_to_be_clickable((By.CLASS_NAME, 'gt_slider_knob')))
            if element:
                return element
        except TimeoutException as e:
            print('超时错误,继续')
            time.sleep(5)


if __name__ == '__main__':
    try:
        count = 6  # 最多识别6次
        driver = webdriver.Chrome()
        # 等待滑动按钮加载完成
        element = main_check_slider(driver)
        while count > 0:
            main_check_code(driver,element)
            time.sleep(2)
            try:
                success_element = (By.CSS_SELECTOR, '.gt_holder .gt_ajax_tip.gt_success')
                # 得到成功标志
                print('suc=',driver.find_element_by_css_selector('.gt_holder .gt_ajax_tip.gt_success'))
                success_images = WebDriverWait(driver, 20).until(EC.presence_of_element_located(success_element))
                if success_images:
                    print('成功识别!!!!!!')
                    count = 0
                    break
            except NoSuchElementException as e:
                print('识别错误,继续')
                count -= 1
                time.sleep(2)
        else:
            print('too many attempt check code ')
            exit('退出程序')
    finally:
        driver.close()

成功识别标志css

CDSY,CDSY.XYZ
方便获取更多学习、工作、生活信息请关注本站微信公众号城东书院 微信服务号城东书院 微信订阅号
推荐内容
相关内容
栏目更新
栏目热门
本栏推荐