解决滑块验证问题（selenium, opencv）
本文环境配置：
系统=>windows10：64位
工具=>PyCharm：2018.1.4
语言=>Python：3.6
第三方库
urllib3
opencv
selenium
PIL
geckodriver.exe（火狐）或 Chromedriver.exe
本次主要是使用selenium模拟滑动验证
最终目标：实现自动点击滑块，并且拖动完成验证
代码实现如下：
import random
from scrapy.http import HtmlResponse
from selenium.common.exceptions import TimeoutException
import time
from selenium import webdriver
from selenium.webdriver.firefox.options import Options
from selenium.webdriver.common.action_chains import ActionChains
from PIL import Image
import cv2
from selenium.webdriver.common.by import By
from selenium.webdriver.support.ui import WebDriverWait
from selenium.webdriver.support import expected_conditions as EC
import numpy as np
import urllib3
from urllib3.connectionpool import xrange
class SeleniumMiddleware(object):
box1 = 49
box2 = 52
convex = 8
cut = 70
image = None
width = 0
high = 0
left = 0
top = 0
leftConvex = 0
topConvex = 0
def __init__(self):
self.options = Options()
self.browser = webdriver.Firefox(executable_path="F:\Python\geckodriver.exe",
firefox_options=self.options)
self.wait = WebDriverWait(self.browser, 20)
# 加载图片
def load(self, path):
self.width = 0
self.high = 0
self.top = 0
self.left = 0
self.leftConvex = 0
self.topConvex = 0
self.image = cv2.imread(path)
if (self.image is None):
return False
h, w, b = self.image.shape
if (w <= 0) or (h <= 0):
return False
self.width = w
self.high = h
return True
# 计算碎片顶边
def getTop(self):
if (self.image is None):
return -1
ckLeft = self.convex * 2
ckHigh = self.high - (self.box1 + self.convex)
# 遍历碎片指定列
j = ckLeft
for i in range(self.convex, ckHigh):
if self.image[i, j][0] != 0 and self.image[i, j][1] != 0 and self.image[i, j][2] != 0:
self.top = i
return i
return self.top
# 计算碎片是否左凸
def getLeftConvex(self):
if (self.image is None):
return False
ckLeft = int(self.convex / 2)
ckHigh = self.top + self.box1
# 遍历碎片指定列
j = ckLeft
for i in range(self.top, ckHigh):
if self.image[i, j][0] != 0 and self.image[i, j][1] != 0 and self.image[i, j][2] != 0:
self.leftConvex = 1
return True
return False
# 计算碎片是否上凸
def getTopConvex(self):
if (self.image is None):
return False
ckLeft = self.convex
ckWidth = ckLeft + self.box1
# 遍历碎片指定列
i = self.top - int(self.convex / 2)
for j in range(ckLeft, ckWidth):
if self.image[i, j][0] != 0 and self.image[i, j][1] != 0 and self.image[i, j][2] != 0:
self.topConvex = 1
return True
return False
def handle(self, path):
if (self.load(path) == False):
return -1
if (self.getTop() == -1):
return -1
self.getLeftConvex()
self.getTopConvex()
# 截取左侧图片
leftImg = self.image[0:self.high, 0:self.cut]
# 高斯模糊
leftImg = cv2.GaussianBlur(leftImg, (3, 3), 0)
# 灰度化
leftImg = cv2.cvtColor(leftImg, cv2.COLOR_BGR2GRAY)
# 二值化
ret, leftImg = cv2.threshold(leftImg, 1, 255, 0)
# 轮廓检测，RETR_EXTERNAL只提取最外层轮廓
tmpImg, leftContours, hierarchy = cv2.findContours(leftImg, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
if (len(leftContours) <= 0):
return -1
# 截取右侧图片
rightImg = self.image[0:self.high, self.cut:self.width]
# 高斯模糊
rightImg = cv2.GaussianBlur(rightImg, (7, 7), 0)
# 灰度化
rightImg = cv2.cvtColor(rightImg, cv2.COLOR_BGR2GRAY)
# 边缘检测
rightImg = cv2.Canny(rightImg, 120, 255)
# 二值化
ret, rightImg = cv2.threshold(rightImg, 100, 255, 0)
# 轮廓检测
tmpImg, rightContours, hierarchy = cv2.findContours(rightImg, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
contoursLen = len(rightContours)
if (contoursLen <= 0):
return -1
# 删除无效轮廓
contours = []
for i in range(contoursLen):
err = 0
# 计算最高坐标
top = self.high
bottom = 0
for j in range(len(rightContours[i])):
y = rightContours[i][j][0][1]
if (y < self.top - 20):
err = 1
break
if (y > self.top + self.box2 + 20):
err = 1
break
if (y < top):
top = y
if (y > bottom):
bottom = y
if (err == 0):
if (self.topConvex):
if (abs(self.top - (top - self.convex)) > 20):
err = 1
else:
if (abs(self.top - top) > 20):
err = 1
if (err == 0):
h = bottom - top
if (h < 40) or (h > 80):
err = 1
if (err == 0):
# 拟合
epsilon = 0.001 * cv2.arcLength(rightContours[i], True)
approx = cv2.approxPolyDP(rightContours[i], epsilon, True)
contours.append(approx)
rightContours = contours
contoursLen = len(rightContours)
# 提取最优轮廓
j = -1
k = 99999
arr = [k for i in range(contoursLen)]
# 面积
leftArea = cv2.contourArea(leftContours[0])
# 周长
leftPerimeter = cv2.arcLength(leftContours[0], True)
for i in range(contoursLen):
# 面积
# rightArea = cv2.contourArea(rightContours[i])
# 周长
# rightPerimeter = cv2.arcLength(rightContours[i], True)
# 轮廓相似度
match = cv2.matchShapes(leftContours[0], rightContours[i], 1, 0.0)
arr[i] = match
# 取差异最小数据
for i in range(contoursLen):
if (arr[i] < k):
k = arr[i]
j = i
# 计算最左位置
self.left = self.width
for i in range(len(rightContours[j])):
if (rightContours[j][i][0][0] < self.left):
self.left = rightContours[j][i][0][0]
if (self.leftConvex):
self.left += (self.convex + 1)
self.left += self.cut
print('距离左边:', (self.left - 14) * 280 / 349)
return (self.left - 14) * 280 / 349
def show(self):
if (self.image is None):
return
x1 = self.left
y1 = self.top
x2 = x1 + self.box1
y2 = y1 + self.box1
temp = self.image.copy()
cv2.rectangle(temp, (x1, y1), (x2, y2), (0, 0, 255), thickness=2)
cv2.imshow("image", temp)
cv2.waitKey(0)
cv2.destroyAllWindows()
def get_track(self, distance):
"""[summary]
根据偏移量获取移动轨迹
Arguments:
distance {[type]} -- 偏移量
"""
distance += 20 # 先滑过一些
# 移动轨迹
track = []
# 当前位移
current = 0
# 减速阀值(达到这个值开始减速)
mid = distance * 4 / 5
# 计算间隔
t = 0.3
# 初速度
v = 0
while current < distance:
if current < mid:
# 加速度为正
a = 3
else:
# 加速度为负
a = -2
# 初速度v0
v0 = v
# 移动距离 x = v0t + 1/2*a*t*t
move = v0 * t + 1 / 2 * a * t * t
# 当前位移
current += move
# 加入轨迹
track.append(round(move))
# 当前速度 v = v0 + at
v = v0 + a * t
# 反着滑动到准确位置
back_tracks = [-3, -3, -2, -2, -2, -2, -2, -1, -1, -1, -1] # 总共等于-20
return {'track': track, 'back_tracks': back_tracks}
def move_to_gap(self, slider, tracks):
"""[summary]
拖动滑块到缺口处
动作： 点击且不释放鼠标-拖拽图片到缺口-释放鼠标
Arguments:
slider {[type]} -- 滑块
tracks {[type]} -- 轨迹
"""
ActionChains(self.browser).move_to_element(slider).perform()
# 点击并按住滑块
ActionChains(self.browser).click_and_hold(slider).perform()
# 移动
for x in tracks['track']:
y = random.randint(-3, 3)
ActionChains(self.browser).move_by_offset(xoffset=x, yoffset=y).perform()
time.sleep(0.6)
for back_track in tracks['back_tracks']:
ActionChains(self.browser).move_by_offset(xoffset=back_track, yoffset=0).perform()
time.sleep(0.3)
ActionChains(self.browser).move_by_offset(xoffset=4, yoffset=0).perform() # 先移过一点
time.sleep(0.9)
ActionChains(self.browser).move_by_offset(xoffset=-4, yoffset=0).perform() # 再退回来
# 释放滑块
time.sleep(0.6)
ActionChains(self.browser).release().perform()
def process_request(self):
try:
# print("url is :::::::::::", request.url)
# print(':::::::::::::::::::::::::::::::')
self.browser.get(
'https://www.anjuke.com/captcha-verify/?callback=shield&from=antispam&serialID=582d45fc4d908d19194f5c8be25b69e5_048c47fa2baa45768dee22e218344755&history=aHR0cHM6Ly9zaGFuZ2hhaS5hbmp1a2UuY29tL2NvbW11bml0eS9qaW5nYW4v')
if '向右滑动滑块' in self.browser.page_source:
print('出现验证码-----------------')
self.browser.execute_script(
"var captcha__puzzleImg = document.evaluate('//*[@class=\"dvc-captcha__puzzleImg\"]', document, null, 0, null).iterateNext();captcha__puzzleImg.style.backgroundColor='black';document.evaluate('//*[@class=\"dvc-operate\"]', document, null, 0, null).iterateNext().style.display='none';")
time.sleep(1)
# print('1111111111111111111111111111111111111')
img = self.browser.find_element_by_css_selector('.info')
img.screenshot('ele.png')
im = Image.open('./ele.png')
# print(im.size)
cro = im.crop((127, 70, 476, 250))
cro.save('./cut.png')
time.sleep(3)
left = self.handle('C:\\Users\lijk_\PycharmProjects\\untitled2\cut.png')
# print(left)
slide = self.wait.until(EC.presence_of_element_located((By.CLASS_NAME, 'dvc-slider__handler')))
track = self.get_track(left)
print(track)
# self.show()
self.move_to_gap(slide, track)
for i in range(7):
if '向右滑动滑块' in self.browser.page_source:
time.sleep(1)
self.move_to_gap(slide, track)
time.sleep(20)
except TimeoutException as e:
print('超时', e)
text = SeleniumMiddleware()
text.process_request()
小结。
其实上面的代码还可以进一步“优化”。例如，当尝试三次滑动后如果仍然没有“验证成功”，就应该主动刷新 使用try进入下一次验证过程。除此之外，以上只是对“滑块验证”进行了分析和模拟，实际情况是，通过了“滑块验证”后，网站监测到是机器模拟会重新出现一个点击验证。关于监测到selenium模拟https://blog.csdn.net/qq_42782937/article/details/86502492
以上代码还有待继续完善，也欢迎看到这篇博文的人多多指正不足之处。