Selenium工具破解滑动验证码（2019.1）

最近跟着书学习爬虫，发现书上关于极致验证码的内容已经过时了，按照书上操作已经得不到期望的效果，于是就自己动手写一个滑动验证码的破解。

工具

：Selenium、python、mongodb

基本思路:
1.使用selenium工具访问b站
2.自动输入用户名和密码
3.获取验证码图片
4.获取带缺口的验证码图片
5.识别缺口位置
6.算法模拟人拖动滑条
考虑到b站自动登录似乎没有什么商业价值，这边就直接上源码了

代码

虽然代码比较另类，但识别原理和网上大部分不可用的例子的识别方法其实差别不大，主要是对验证码的位置进行了修正，另外在拖动滑条时使用模拟人的校准行为，使得自动登录可用，最后用Mongodb记录成功与失败次数

from selenium import webdriver
import time
from io import BytesIO
from PIL import Image
from selenium.webdriver.common.action_chains import ActionChains
import pymongo


#browser = webdriver.Chrome()
client = pymongo.MongoClient(host='localhost', port=27017)    # 建立连接
db = client.test    # 指定数据库
collection = db.login_bilibili    # 指定表


global success
global failed


def counts():
    global success
    global failed
    success = 0
    failed = 0


def get_img():
    scr = browser.get_screenshot_as_png()
    scr = Image.open(BytesIO(scr))
    img = browser.find_element_by_class_name('gt_box')
    scr_locate = img.location
    size = img.size
    print('imgsize:',size)
    top, bottom, left, right=scr_locate['y']+60, scr_locate['y']+80 + size['height'],scr_locate['x']+150, scr_locate['x'] + size['width']+180
    scr_img = scr.crop((left, top,right, bottom))
    #scr_img.show()
    return scr_img


def difference(img1, img2, x, y):
    """两张图片每个像素点对比，差值过大的为拼图位置"""
    p1 = img1.load()[x, y]
    p2 = img2.load()[x, y]
    p = 60
    #print('p1:',p1,'\np2:',p2)
    if abs(p1[0]-p2[0]) < p and abs(p1[1]-p2[1]) < p and abs(p1[2]-p2[2]) < p:
        return True
    else:
        print(p1, p2)
        return False


def get_crave(img1, img2):
    ahead = 80
    for i in range(ahead, img1.size[0]):
        for j in range(img1.size[1]):
            if not difference(img1, img2, i, j):
                print('第%s列' % i)
                crave = i
                return crave


def get_track(s):
    track = []
    for t in range(36):
        if t < 21:
            t = t/20
            a1 = 8*s/5
            v0 = a1 * t
            t = 0.05
            s1 = v0 * t + 0.5 * a1 * t * t
            track.append(s1)
        else:
            a2 = -27*s/8
            v0 = 8*s/5 + a2 * (-20 + t)/20
            t = 0.05
            s2 = v0*t + 0.5 * a2 * t * t
            track.append(s2)
    return track


def login(user, pwd):
    global failed
    browser.get('https://passport.bilibili.com/login')  # 输入网址
    time.sleep(1.5)
    username = browser.find_element_by_id('login-username')  # 定位id为draggable的节点
    username.send_keys(user)
    passwd = browser.find_element_by_id('login-passwd')  # 定位id为draggable的节点
    passwd.send_keys(pwd)
    button = browser.find_element_by_class_name("gt_slider_knob")
    ActionChains(browser).move_to_element(button).perform()  # 鼠标移动到滑块上不点击
    time.sleep(0.7)
    img1 = get_img()  # 没有缺口的图
    ActionChains(browser).click_and_hold(browser.find_element_by_class_name("gt_slider_knob")).perform()  # 按住不放
    img2 = get_img()  # 有缺口的图
    crave = get_crave(img1, img2) - 10
    if crave < 85:
        browser.quit()
        failed += 1
        return crave + 10
    if crave < 135:
        crave += 4
    if crave > 135 and crave < 189:
        crave -= 5
    if crave >= 190:
        crave -= 19
    print('carve:', crave)
    move(crave)
    time.sleep(3)
    browser.quit()
    return ''


def move(crave):
    global success
    s = 0
    j = 0
    track = get_track(crave)
    if crave < 100:
        track.extend([-3, 2.5, 1.5, -1.5, 1.5, -2.5])
    else:
        track.extend([-3, 4.1, -4.1, 1.5, -1.5])
    for i in track:
        j += 1
        if j > 36:
            time.sleep(0.5)
        ActionChains(browser).move_by_offset(i, 0).perform()
        time.sleep(0.05)
        s += i
    print('移动:', s)
    ActionChains(browser).release(browser.find_element_by_class_name("gt_slider_knob")).perform()
    time.sleep(2)
    if browser.current_url == 'https://www.bilibili.com/':
        success += 1



counts()
if __name__ == '__main__':
    while True:
        try:
            browser = webdriver.Chrome()
            failed_crave = login(用户名, 密码)
            print('成功%s次，失败%s次' % (success, failed))
            total = {
                'id': 1,
                '登陆成功次数': success,
                '登陆失败次数': failed,
                '成功率': success / (success + failed),
                'failed_crave': failed_crave
            }
            r = collection.insert_one(total)
        except:
            failed += 1
            total = {
                'id': 1,
                '登陆成功次数': success,
                '登陆失败次数': failed,
                '成功率': success / (success + failed),
                'failed_crave': '此次是字母数字验证码'
            }
            r = collection.insert_one(total)

测试

经过挂机一晚上的测试，成功率在70%以上，但美中不足的是，在访问300次左右后，错误率开始上升，分析应该是访问过于频繁导致出现了特殊错误，到500次以后的数据基本上全都是失败