使用OpenCVselenium解决滑块验证码

Posted 2021-08-24 GuanLu

本次案例使用OpenCV和selenium来解决一下滑块验证码

先说一下思路：

1. 弹出滑块验证码后使用selenium元素截图将验证码整个背景图截取出来
2. 将需要滑动的小图单独截取出来，最好将小图与背景图顶部的像素距离获取到，这样可以将背景图上下多余的边框截取掉
3. 使用OpenCV将背景图和小图进行灰度处理，并对小图再次进行二值化全局阈值，这样就可以利用OpenCV在背景图中找到小图所在的位置
4. 用OpenCV获取到相差的距离后利用selenium的鼠标拖动方法进行拖拉至终点。

 

详细步骤：

先获取验证码背景图，selenium浏览器对象中使用screenshot方法可以将指定的元素图片截取出来

import os
from selenium import webdriver


browser = webdriver.Chrome()
browser.get("https://www.toutiao.com/c/user/token/MS4wLjABAAAA4EKNlqVeNTTuEdWn0VytNS8cdODKTsNNwLTxOnigzZtclro2Kylvway5mTyTUKvz/")

save_path = os.path.join(os.path.expanduser(\'~\'), "Desktop", "background.png")
browser.find_element_by_id("element_id_name").screenshot(save_path)

 

截取后的验证码背景图和需要滑动的小图   如：

   

再将小图与背景图顶部的像素距离获取到，指的是下面图中红边的高度：

如果html元素中小图是单独存在时，那么它的高度在会定义在页面元素中，使用selenium页面元素对象的value_of_css_property方法可以获取到像素距离。

获取这个是因为要把背景图的上下两边多余部分进行切除，从而保留关键的图像部位，能够大幅度提高识别率。

element_object = browser.find_element_by_xpath("xpath_element")
px = element_object.value_of_css_property("top")

 

接下来就要对图像进行灰度处理：

import numpy
import cv2


def make_threshold(img):
    """全局阈值
    将图片二值化,去除噪点,让其黑白分明"""
    x = numpy.ones(img.shape, numpy.uint8) * 255
    y = img - x
    result, thresh = cv2.threshold(y, 127, 255, cv2.THRESH_BINARY_INV)
    # 将二值化后的结果返回
    return thresh


class ComputeDistance:
    """获取需要滑动的距离
    将验证码背景大图和需要滑动的小图进行处理,先在大图中找到相似的小图位置,再获取对应的像素偏移量"""
    def __init__(self, Background_path: str, image_to_move: str, offset_top_px: int):
        """
        :param Background_path: 验证码背景大图
        :param image_to_move: 需要滑动的小图
        :param offset_top_px: 小图距离在大图上的顶部边距(像素偏移量)
        """
        self.Background_img = cv2.imread(Background_path)
        self.offset_px = offset_top_px
        self.show_img = show_img
        small_img_data = cv2.imread(image_to_move, cv2.IMREAD_UNCHANGED)
        # 得到一个改变维度为50的乘以值
        scaleX = 50 / small_img_data.shape[1]
        # 使用最近邻插值法缩放，让xy乘以scaleX，得到缩放后shape为50x50的图片
        self.tpl_img = cv2.resize(small_img_data, (0, 0), fx=scaleX, fy=scaleX)
        self.Background_cutting = None

    def tpl_op(self):
        # 将小图转换为灰色
        tpl_gray = cv2.cvtColor(self.tpl_img, cv2.COLOR_BGR2GRAY)
        h, w = tpl_gray.shape
        # 将背景图转换为灰色
        # Background_gray = cv2.cvtColor(self.Background_img, cv2.COLOR_BGR2GRAY)
        Background_gray = cv2.cvtColor(self.Background_cutting, cv2.COLOR_BGR2GRAY)
        # 得到二值化后的小图
        threshold_img = make_threshold(tpl_gray)
        # 将小图与大图进行模板匹配,找到所对应的位置
        result = cv2.matchTemplate(Background_gray, threshold_img, cv2.TM_CCOEFF_NORMED)
        min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(result)
        # 左上角位置
        top_left = (max_loc[0] - 5, max_loc[1] + self.offset_px)
        # 右下角位置
        bottom_right = (top_left[0] + w, top_left[1] + h)
        # 在源颜色大图中画出小图需要移动到的终点位置
        """rectangle(图片源数据, 左上角, 右下角, 颜色, 画笔厚度)"""
        cv2.rectangle(self.Background_img, top_left, bottom_right, (0, 0, 255), 2)

    def cutting_background(self):
        """切割图片的上下边框"""
        height = self.tpl_img.shape[0]
        # 将大图中上下多余部分去除,如: Background_img[40:110, :]
        self.Background_cutting = self.Background_img[self.offset_px - 10: self.offset_px + height + 10, :]

    def run(self):
        # 如果小图的长度与大图的长度一致则不用将大图进行切割,可以将self.cutting_background()注释掉
        self.cutting_background()
        return self.tpl_op()


if __name__ == \'__main__\':
    image_path1 = "背景图路径"
    image_path2 = "小图路径"
    distance_px = "像素距离"
    main = ComputeDistance(image_path1, image_path2, distance_px)
    main.run()

 

上面代码可以返回小图到凹点的距离，现在我们可以看一下灰度处理中的图片样子：

     

 

得到距离后还要对这个距离数字进行处理一下，要让它拆分成若干个小数，这么做的目的是在拖动的时候不能一下拖动到终点，

要模仿人类的手速缓缓向前行驶，不然很明显是机器在操控。

比如到终点的距离为100，那么要把它转为 [8, 6, 11, 10, 3, 6, 3, -2, 4, 0, 15, 1, 9, 6, -2, 4, 1, -2, 15, 6, -2] 类似的，列表中的数加起来正好为100.

最简单的转换：

def handle_distance(distance):
    """将直线距离转为缓慢的轨迹"""
    import random
    slow_distance = []
    while sum(slow_distance) <= distance:
        slow_distance.append(random.randint(-2, 15))

    if sum(slow_distance) != distance:
        slow_distance.append(distance - sum(slow_distance))
    return slow_distance

 

有了到终点的距离，接下来就开始拖动吧：

import time
from random import randint
from selenium.webdriver.common.action_chains import ActionChains


def move_slider(website, slider, track, **kwargs):
    """将滑块移动到终点位置
    :param website: selenium页面对象
    :param slider: selenium页面中滑块元素对象
    :param track: 到终点所需的距离
    """
    name = kwargs.get(\'name\', \'滑块\')

    try:
        if track[0] > 200:
            return track[0]
        # 点击滑块元素并拖拽
        ActionChains(website).click_and_hold(slider).perform()
        time.sleep(0.15)
        for i in track:
            # 随机上下浮动鼠标
            ActionChains(website).move_by_offset(xoffset=i, yoffset=randint(-2, 2)).perform()
        # 释放元素
        time.sleep(1)
        ActionChains(website).release(slider).perform()
        time.sleep(1)
        # 随机拿开鼠标
        ActionChains(website).move_by_offset(xoffset=randint(200, 300), yoffset=randint(200, 300)).perform()
        print(f\'[网页] 拖拽 {name}\')
        return True
    except Exception as e:
        print(f\'[网页] 拖拽 {name} 失败 {e}\')
        

 

 

 

————————————————————————————————————————————————————

教程结束，让我们结合上面代码做一个案例吧。

调用ComputeDistance类时，参数 show_img=True 可以在拖动验证码前进行展示背景图识别终点后的区域在哪里， 如：

distance_obj = ComputeDistance(background_path, small_path, px, show_img=True)

 

OK，下面为案例代码：

import os
import time
import requests
import cv2
import numpy
from random import randint

from selenium import webdriver
from selenium.webdriver.common.action_chains import ActionChains


def show_image(img_array, name=\'img\', resize_flag=False):
    """展示图片"""
    maxHeight = 540
    maxWidth = 960
    scaleX = maxWidth / img_array.shape[1]
    scaleY = maxHeight / img_array.shape[0]
    scale = min(scaleX, scaleY)
    if resize_flag and scale < 1:
        img_array = cv2.resize(img_array, (0, 0), fx=scale, fy=scale)
    cv2.imshow(name, img_array)
    cv2.waitKey(0)
    cv2.destroyWindow(name)


def make_threshold(img):
    """全局阈值
    将图片二值化,去除噪点,让其黑白分明"""
    x = numpy.ones(img.shape, numpy.uint8) * 255
    y = img - x
    result, thresh = cv2.threshold(y, 127, 255, cv2.THRESH_BINARY_INV)
    # 将二值化后的结果返回
    return thresh


def move_slider(website, slider, track, **kwargs):
    """将滑块移动到终点位置
    :param website: selenium页面对象
    :param slider: selenium页面中滑块元素对象
    :param track: 到终点所需的距离
    """
    name = kwargs.get(\'name\', \'滑块\')

    try:
        if track[0] > 200:
            return track[0]
        # 点击滑块元素并拖拽
        ActionChains(website).click_and_hold(slider).perform()
        time.sleep(0.15)
        for i in track:
            # 随机上下浮动鼠标
            ActionChains(website).move_by_offset(xoffset=i, yoffset=randint(-2, 2)).perform()
        # 释放元素
        time.sleep(1)
        ActionChains(website).release(slider).perform()
        time.sleep(1)
        # 随机拿开鼠标
        ActionChains(website).move_by_offset(xoffset=randint(200, 300), yoffset=randint(200, 300)).perform()
        print(f\'[网页] 拖拽 {name}\')
        return True
    except Exception as e:
        print(f\'[网页] 拖拽 {name} 失败 {e}\')


class ComputeDistance:
    """获取需要滑动的距离
    将验证码背景大图和需要滑动的小图进行处理,先在大图中找到相似的小图位置,再获取对应的像素偏移量"""
    def __init__(self, Background_path: str, image_to_move: str, offset_top_px: int, show_img=False):
        """
        :param Background_path: 验证码背景大图
        :param image_to_move: 需要滑动的小图
        :param offset_top_px: 小图距离在大图上的顶部边距(像素偏移量)
        :param show_img: 是否展示图片
        """
        self.Background_img = cv2.imread(Background_path)
        self.offset_px = offset_top_px
        self.show_img = show_img
        small_img_data = cv2.imread(image_to_move, cv2.IMREAD_UNCHANGED)
        # 得到一个改变维度为50的乘以值
        scaleX = 50 / small_img_data.shape[1]
        # 使用最近邻插值法缩放，让xy乘以scaleX，得到缩放后shape为50x50的图片
        self.tpl_img = cv2.resize(small_img_data, (0, 0), fx=scaleX, fy=scaleX)
        self.Background_cutting = None

    def show(self, img):
        if self.show_img:
            show_image(img)

    def tpl_op(self):
        # 将小图转换为灰色
        tpl_gray = cv2.cvtColor(self.tpl_img, cv2.COLOR_BGR2GRAY)
        h, w = tpl_gray.shape
        # 将背景图转换为灰色
        # Background_gray = cv2.cvtColor(self.Background_img, cv2.COLOR_BGR2GRAY)
        Background_gray = cv2.cvtColor(self.Background_cutting, cv2.COLOR_BGR2GRAY)
        # 得到二值化后的小图
        threshold_img = make_threshold(tpl_gray)
        # 将小图与大图进行模板匹配,找到所对应的位置
        result = cv2.matchTemplate(Background_gray, threshold_img, cv2.TM_CCOEFF_NORMED)
        min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(result)
        # 左上角位置
        top_left = (max_loc[0] - 5, max_loc[1] + self.offset_px)
        # 右下角位置
        bottom_right = (top_left[0] + w, top_left[1] + h)
        # 在源颜色大图中画出小图需要移动到的终点位置
        """rectangle(图片源数据, 左上角, 右下角, 颜色, 画笔厚度)"""
        cv2.rectangle(self.Background_img, top_left, bottom_right, (0, 0, 255), 2)
        if self.show_img:
            show_image(self.Background_img)
        return top_left

    def cutting_background(self):
        """切割图片的上下边框"""
        height = self.tpl_img.shape[0]
        # 将大图中上下多余部分去除,如: Background_img[40:110, :]
        self.Background_cutting = self.Background_img[self.offset_px - 10: self.offset_px + height + 10, :]

    def run(self):
        # 如果小图的长度与大图的长度一致则不用将大图进行切割,可以将self.cutting_background()注释掉
        self.cutting_background()
        return self.tpl_op()


class TodayNews(object):
    def __init__(self):
        self.url = "https://www.toutiao.com/c/user/token/" \\
                   "MS4wLjABAAAA4EKNlqVeNTTuEdWn0VytNS8cdODKTsNNwLTxOnigzZtclro2Kylvway5mTyTUKvz/"
        self.process_folder = os.path.join(os.path.expanduser(\'~\'), "Desktop", "today_news")
        self.background_path = os.path.join(self.process_folder, "background.png")
        self.small_path = os.path.join(self.process_folder, "small.png")
        self.small_px = None
        self.xpath = {}
        self.browser = None

    def check_file_exist(self):
        """检查流程目录是否存在"""
        if not os.path.isdir(self.process_folder):
            os.mkdir(self.process_folder)

    def start_browser(self):
        """启动浏览器"""
        self.browser = webdriver.Chrome()
        self.browser.maximize_window()

    def close_browser(self):
        self.browser.quit()

    def wait_element_loaded(self, xpath: str, timeout=10, close_browser=True):
        """等待页面元素加载完成
        :param xpath: xpath表达式
        :param timeout: 最长等待超时时间
        :param close_browser: 元素等待超时后是否关闭浏览器
        :return: Boolean
        """
        now_time = int(time.time())
        while int(time.time()) - now_time < timeout:
            # noinspection PyBroadException
            try:
                element = self.browser.find_element_by_xpath(xpath)
                if element:
                    return True
                time.sleep(1)
            except Exception:
                pass
        else:
            if close_browser:
                self.close_browser()
            # print("查找页面元素失败，如果不存在网络问题请尝试修改xpath表达式")
            return False

    def add_page_element(self):
        self.xpath[\'background_img\'] = \'//div[@role="dialog"]/div[2]/img[1]\'
        self.xpath[\'small_img\'] = \'//div[@role="dialog"]/div[2]/img[2]\'
        self.xpath[\'slider_button\'] = \'//div[@id="secsdk-captcha-drag-wrapper"]/div[2]\'

    def process_main(self):
        """处理页面内容"""
        self.browser.get(self.url)

        for _ in range(10):
            if self.wait_element_loaded(self.xpath[\'background_img\'], timeout=5, close_browser=False):
                time.sleep(1)
                # 截图
                self.browser.find_element_by_xpath(self.xpath[\'background_img\']).screenshot(self.background_path)
                small_img = self.browser.find_element_by_xpath(self.xpath[\'small_img\'])
                # 获取小图片的URL链接
                small_url = small_img.get_attribute("src")
                # 获取小图片距离背景图顶部的像素距离
                self.small_px = small_img.value_of_css_property("top").replace("px", "").split(".")[0]

                response = requests.get(small_url)
                if response.ok:
                    with open(self.small_path, "wb") as file:
                        file.write(response.content)

                time.sleep(1)
                # 如果没滑动成功则刷新页面重试
                if not self.process_slider():
                    self.browser.refresh()
                    continue
            else:
                break

    @staticmethod
    def handle_distance(distance):
        """将直线距离转为缓慢的轨迹"""
        import random
        slow_distance = []
        while sum(slow_distance) <= distance:
            slow_distance.append(random.randint(-2, 15))

        if sum(slow_distance) != distance:
            slow_distance.append(distance - sum(slow_distance))
        return slow_distance

    def process_slider(self):
        """处理滑块验证码"""

        distance_obj = ComputeDistance(self.background_path, self.small_path, int(self.small_px), show_img=False)
        # 获取移动所需的距离
        distance = distance_obj.run()

        track = self.handle_distance(distance[0])
        track.append(-2)
        slider_element = self.browser.find_element_by_xpath(self.xpath[\'slider_button\'])

        move_slider(self.browser, slider_element, track)
        time.sleep(2)

        # 如果滑动完成则返回True
        if not self.wait_element_loaded(self.xpath[\'slider_button\'], timeout=2, close_browser=False):
            return True
        else:
            return False

    def run(self):
        self.check_file_exist()
        self.start_browser()
        self.add_page_element()
        self.process_main()
        # self.close_browser()


if __name__ == \'__main__\':
    main = TodayNews()
    main.run()