爬虫与反爬-旋转验证码突破方案（知名短视频、TK海外版及某东等等）

概述：文本对旋转验证码进行了突破及讲述了实现原理，代码使用纯算法 OpenCV，使用代价较小同时不用安装一大堆AI训练相关的模组，方便且能够快速上手

当前亲自验证了能够支持的网站：国内知名短视频平台、海外版以及某东等等使用了旋转验证码的网站

基本上是旋转验证码的网站都能过，测试成功率在 99% 以上

实现步骤：

一、下载图片

为了兼容，在这里可以支持自己传base64，或者自己传链接来下载图片

def request_image_content(image_url, proxies: Optional[dict] = None):
    """
    图片下载
    :param image_url: 
    :param proxies: 
    :return: 
    """
    import requests  # type: ignore

    headers = {
        "user-agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 "
        "(KHTML, like Gecko) Chrome/108.0.0.0 Safari/537.36"
    }
    response = requests.get(image_url, headers=headers, proxies=proxies)
    return response.content



def image_to_cv2(base_image: str, image_type: int, grayscale: bool, proxies=None):
    """
    根据 图片类型来读取图片数据
    :param base_image: 
    :param image_type: 
    :param grayscale: 
    :param proxies: 
    :return: 
    """
    assert image_type in [0, 1, 2]
    if image_type == 0:
        search_base64 = re.search("base64,(.*?)$", base_image)
        base64_image = search_base64.group(1) if search_base64 else base_image
        image_array = np.asarray(
            bytearray(base64.b64decode(base64_image)), dtype="uint8"
        )
        image = cv2.imdecode(image_array, cv2.IMREAD_COLOR)
    elif image_type == 1:
        image_content = request_image_content(base_image, proxies)
        if not image_content:
            raise Exception("请求图片链接失败！")
        image_array = np.array(bytearray(image_content), dtype=np.uint8)
        image = cv2.imdecode(image_array, cv2.IMREAD_COLOR)
    else:
        image = cv2.imread(base_image)

    if grayscale:
        image = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)

    return image

二、识别图像像素相似度

核心原理：使用opencv对两张图片进行合并，同时从0度开始，一点点旋转，选择像素相似度最高的一个度数

步骤1：将外圈图与内圈图的边缘像素进行匹配，获取相似度

            mat_rotate = cv2.getRotationMatrix2D((h * 0.5, w * 0.5), rtype * angle, 1)
            dst = cv2.warpAffine(inner_image, mat_rotate, (h, w))
            ret = cv2.matchTemplate(outer_image, dst, cv2.TM_CCOEFF_NORMED)
            similar_value = cv2.minMaxLoc(ret)[1]

步骤2：将相似度最高的一个结果进行返回并输出

step作为步长，设置1则1度1度的旋转，会更加精准

    step = 1
    rotate_info_list = [RotateData(0, 0, 1, 361, step)]
    for item in rotate_info_list:
        min_similar_rotate_info = item
        for angle in range(*item[2:]):
            mat_rotate = cv2.getRotationMatrix2D((h * 0.5, w * 0.5), rtype * angle, 1)
            dst = cv2.warpAffine(inner_image, mat_rotate, (h, w))
            ret = cv2.matchTemplate(outer_image, dst, cv2.TM_CCOEFF_NORMED)
            similar_value = cv2.minMaxLoc(ret)[1]
            if similar_value < min_similar_rotate_info.similar:
                continue
            rotate_info = RotateData(similar_value, angle, angle - step, angle + step, step)
            rotate_info_list.append(rotate_info)
            if len(rotate_info_list) > 5:
                rotate_info_list.remove(min_similar_rotate_info)
            min_similar_rotate_info = min(rotate_info_list)