[Python]爬虫获取知乎某个问题下所有图片并去除水印

获取URL

进入某个知乎问题的主页下，按F12打开开发者工具后查看network面板。

network面板可以查看页面向服务器请求的资源、资源的大小、加载资源花费的时间以及哪些资源加载失败等信息。还可以查看HTTP的请求头，返回内容等。

以“你有哪些可爱的猫猫照片？”问题为例，我们可以看到network面板如下：

按一下快捷键Ctrl + F在搜索面板中直接搜索对应的答案出现的文字，可以找到对应的目标url及其response：

安装对应的package，其他包都比较简单，需要注意的是python图像处理的包cv2安装命令如下：

pip install opencv-python

URL分析

1. 参数分析

我们刚才获取的URL如下：

https://www.zhihu.com/api/v4/questions/356541789/answers?include=data%5B*%5D.is_normal%2Cadmin_closed_comment%2Creward_info%2Cis_collapsed%2Cannotation_action%2Cannotation_detail%2Ccollapse_reason%2Cis_sticky%2Ccollapsed_by%2Csuggest_edit%2Ccomment_count%2Ccan_comment%2Ccontent%2Ceditable_content%2Cattachment%2Cvoteup_count%2Creshipment_settings%2Ccomment_permission%2Ccreated_time%2Cupdated_time%2Creview_info%2Crelevant_info%2Cquestion%2Cexcerpt%2Cis_labeled%2Cpaid_info%2Cpaid_info_content%2Crelationship.is_authorized%2Cis_author%2Cvoting%2Cis_thanked%2Cis_nothelp%2Cis_recognized%3Bdata%5B*%5D.mark_infos%5B*%5D.url%3Bdata%5B*%5D.author.follower_count%2Cvip_info%2Cbadge%5B*%5D.topics%3Bdata%5B*%5D.settings.table_of_content.enabled&offset=&limit=3&sort_by=default&platform=desktop

其中包含的参数为：

• limit: 一页显示的答案条数
• offset：页面的偏移量
• sort_by：答案的排序方式，支持默认排序或者按时间排序

2.解析Response

尝试着发一个请求并截获http response：

# python3
import requests
import json

if __name__ == '__main__':
    target_url = "https://www.zhihu.com/api/v4/questions/356541789/answers?include=data%5B*%5D.is_normal%2Cadmin_closed_comment%2Creward_info%2Cis_collapsed%2Cannotation_action%2Cannotation_detail%2Ccollapse_reason%2Cis_sticky%2Ccollapsed_by%2Csuggest_edit%2Ccomment_count%2Ccan_comment%2Ccontent%2Ceditable_content%2Cattachment%2Cvoteup_count%2Creshipment_settings%2Ccomment_permission%2Ccreated_time%2Cupdated_time%2Creview_info%2Crelevant_info%2Cquestion%2Cexcerpt%2Cis_labeled%2Cpaid_info%2Cpaid_info_content%2Crelationship.is_authorized%2Cis_author%2Cvoting%2Cis_thanked%2Cis_nothelp%2Cis_recognized%3Bdata%5B*%5D.mark_infos%5B*%5D.url%3Bdata%5B*%5D.author.follower_count%2Cvip_info%2Cbadge%5B*%5D.topics%3Bdata%5B*%5D.settings.table_of_content.enabled&offset=&limit=3&sort_by=default&platform=desktop"
    headers = {
        'user-agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/72.0.3626.119 Safari/537.36',
    }
    
    response = requests.get(url = target_url, headers = headers)
    html = response.text 
    print(html)

获取到的response如下，我们需要做的是找到所有图片对应的链接，使用Json工具解析后可以从http返回值json中找到图片所在的位置，后续就是通过爬虫解析到下载地址即可：

Tips：值得注意的是网站的返回值样式经常变动，而且不同网站返回值的组织样式也不一样，所以不可盲目借鉴。

3.获取所有答案url

仍然使用在“开发者工具中”查找答案关键字的方法，我们可以拿到多个答案对应的url，我们需要从这些url中找到规律：

https://www.zhihu.com/api/v4/questions/356541789/answers?include=data%5B%2A%5D.is_normal%2Cadmin_closed_comment%2Creward_info%2Cis_collapsed%2Cannotation_action%2Cannotation_detail%2Ccollapse_reason%2Cis_sticky%2Ccollapsed_by%2Csuggest_edit%2Ccomment_count%2Ccan_comment%2Ccontent%2Ceditable_content%2Cattachment%2Cvoteup_count%2Creshipment_settings%2Ccomment_permission%2Ccreated_time%2Cupdated_time%2Creview_info%2Crelevant_info%2Cquestion%2Cexcerpt%2Cis_labeled%2Cpaid_info%2Cpaid_info_content%2Crelationship.is_authorized%2Cis_author%2Cvoting%2Cis_thanked%2Cis_nothelp%2Cis_recognized%3Bdata%5B%2A%5D.mark_infos%5B%2A%5D.url%3Bdata%5B%2A%5D.author.follower_count%2Cvip_info%2Cbadge%5B%2A%5D.topics%3Bdata%5B%2A%5D.settings.table_of_content.enabled&limit=3&offset=3&platform=desktop&sort_by=default

https://www.zhihu.com/api/v4/questions/356541789/answers?include=data%5B%2A%5D.is_normal%2Cadmin_closed_comment%2Creward_info%2Cis_collapsed%2Cannotation_action%2Cannotation_detail%2Ccollapse_reason%2Cis_sticky%2Ccollapsed_by%2Csuggest_edit%2Ccomment_count%2Ccan_comment%2Ccontent%2Ceditable_content%2Cattachment%2Cvoteup_count%2Creshipment_settings%2Ccomment_permission%2Ccreated_time%2Cupdated_time%2Creview_info%2Crelevant_info%2Cquestion%2Cexcerpt%2Cis_labeled%2Cpaid_info%2Cpaid_info_content%2Crelationship.is_authorized%2Cis_author%2Cvoting%2Cis_thanked%2Cis_nothelp%2Cis_recognized%3Bdata%5B%2A%5D.mark_infos%5B%2A%5D.url%3Bdata%5B%2A%5D.author.follower_count%2Cvip_info%2Cbadge%5B%2A%5D.topics%3Bdata%5B%2A%5D.settings.table_of_content.enabled&limit=3&offset=0&platform=desktop&sort_by=default

尽管url的格式不尽相同，但是我发现基本都遵循如下格式，只需要变更offset参数即可

https://www.zhihu.com/api/v4/questions/356541789/answers?include=data%5B%2A%5D.is_normal%2Cadmin_closed_comment%2Creward_info%2Cis_collapsed%2Cannotation_action%2Cannotation_detail%2Ccollapse_reason%2Cis_sticky%2Ccollapsed_by%2Csuggest_edit%2Ccomment_count%2Ccan_comment%2Ccontent%2Ceditable_content%2Cattachment%2Cvoteup_count%2Creshipment_settings%2Ccomment_permission%2Ccreated_time%2Cupdated_time%2Creview_info%2Crelevant_info%2Cquestion%2Cexcerpt%2Cis_labeled%2Cpaid_info%2Cpaid_info_content%2Crelationship.is_authorized%2Cis_author%2Cvoting%2Cis_thanked%2Cis_nothelp%2Cis_recognized%3Bdata%5B%2A%5D.mark_infos%5B%2A%5D.url%3Bdata%5B%2A%5D.author.follower_count%2Cvip_info%2Cbadge%5B%2A%5D.topics%3Bdata%5B%2A%5D.settings.table_of_content.enabled&limit=3&offset=0&platform=desktop&sort_by=default

Code

1. 模拟请求

简单加上headers即可，知乎的校验没有其他网站来得严格，访问过于频繁时会限制访问一段时间，我这里简单使用随机请求头和代理IP来处理：

def get_http_content(number, offset):
    """读取知乎某问题下的答案url, 返回对应json
    Args:
        number: 知乎问题唯一标识
        offset: 偏移量
    """
    target_url = "https://www.zhihu.com/api/v4/questions/{number}/answers?include=data%5B*%5D.is_normal%2Cadmin_closed_comment%2" \
        "Creward_info%2Cis_collapsed%2Cannotation_action%2Cannotation_detail%2Ccollapse_reason%2Cis_sticky%2Ccollapsed_by%2" \
        "Csuggest_edit%2Ccomment_count%2Ccan_comment%2Ccontent%2Ceditable_content%2Cvoteup_count%2Creshipment_settings%2Ccomment_permission%2" \
        "Ccreated_time%2Cupdated_time%2Creview_info%2Crelevant_info%2Cquestion%2Cexcerpt%2Crelationship.is_authorized%2Cis_author%2Cvoting%2" \
        "Cis_thanked%2Cis_nothelp%2Cis_labeled%2Cis_recognized%2Cpaid_info%2Cpaid_info_content%3Bdata%5B*%5D.mark_infos%5B*%5D.url%3Bdata%5B*%5" \
        "D.author.follower_count%2Cbadge%5B*%5D.topics&offset={offset}&limit={limit}&sort_by=default&platform=desktop".format(
            number=number, offset=offset, limit=limit)
    logger.info("target_url:{}", target_url)
    headers = {
        'User-Agent': fake_useragent.get_random_useragent(),
    }
    ip = IPPool().get_random_key()
    proxies = {"http": "http://" + ip}
    response = requests.get(target_url, headers=headers, proxies=proxies)
    if (response is None) or (response.status_code != 200):
        logger.warning("http response is None, number={}, offset={}, status_code={}".format(
            number, offset, response.status_code))
        return None
    html = response.text
    return json.loads(html)

2. 解析出图片地址

def start_crawl():
    """开始爬虫获取图片
    """
    for i in range(0, max_pages):
        offset = limit * i
        logger.info("download pictures with offset {}". format(offset))

        # 获取html
        content_dict = get_http_content(number, offset)
        if content_dict is None:
            logger.error(
                "get http resp fail, number={} offset={}", number, offset)
            continue
        # content_dict['data']存储了答案列表
        if 'data' not in content_dict:
            logger.error("parse data from http resp fail, dict={}", dict)
            continue
        for answer_text in content_dict['data']:
            logger.info(
                "get pictures from answer: https://www.zhihu.com/question/{}/answer/{}", number, answer_text['id'])
            if 'content' not in answer_text:
                logger.error(
                    "parse content from answer text fail, text={}", answer_text)
                continue
            answer_content = pq(answer_text['content'])
            img_urls = answer_content.find('noscript').find('img')
            # 此篇问答不包含图片时打印对应信息, 方便debug
            if len(list(img_urls)) <= 0:
                logger.warning(
                    "this answer has no pictures, url:https://www.zhihu.com/question/{}/answer/{}", number, answer_text['id'])
                continue
            for img_url in img_urls.items():
                # src例子: https://pic2.zhimg.com/50/v2-c970108cd260ea095383627362c1d04f_720w.jpg?source=1940ef5c
                src = img_url.attr("src")
                # 解析出图片格式后缀: .jpeg 或者 .gif等
                source_index = src.rfind('?source')
                if source_index == -1:
                    logger.error("find source index fail, src:{} source_index{}",
                                 src, source_index)
                suffix = src[0:source_index]
                suffix_index = src.rfind('.')
                if source_index == -1:
                    logger.error("find suffix fail, src:{} suffix_index{}".format(
                        src, suffix_index))
                suffix = suffix[suffix_index:]
                logger.info("get picture url, src:{} suffix:{}", src, suffix)
                store_picture(src, suffix)
                time.sleep(1)

3. 将图片存储到本地

def store_picture(img_url, suffix):
    """将图片存储到文件夹中
    Args:
        img_url: 图片链接
        suffix: 图片后缀, 比如'.jpg', '.gif'等
    """
    headers = {
        'User-Agent': fake_useragent.get_random_useragent(),
    }
    ip = IPPool().get_random_key()
    proxies = {"http": "http://" + ip}
    http_resp = requests.get(img_url, headers=headers, proxies=proxies)
    if (http_resp is None) or (http_resp.status_code != 200):
        logger.warning("get http resp fail, url={} http_resp={}",
                       img_url, http_resp)
        return
    content = http_resp.content
    with open(f"{picture_path}/{uuid.uuid4()}{suffix}", 'wb') as f:
        f.write(content)

4. 去除图片水印

本来打算使用图像识别进行抠图去除水印的（因为知乎的水印比较简单而且样式统一），无奈最近需要处理的事情比较多，因此就简单通过opencv包进行裁剪：

def crop_watermark(ori_dir, adjusted_dir):
    """通过裁剪图片的方式来去除水印, 注意无法处理gif格式的图片
    Args:
        ori_dir: 图片所在的文件夹
        adjusted_dir: 去除水印后存放的文件夹
    """
    img_path_list = os.listdir(ori_dir)  # 获取目录下的所有文件
    total = len(img_path_list)
    cnt = 1
    for img_path in img_path_list:
        logger.info(
            "the overall process::{}/{}, now handle the picture:{}", cnt, total, img_path)
        img_abs_path = ori_dir + '/' + img_path
        img = cv2.imread(img_abs_path)
        if img is None:
            logger.error("cv2.imread fail, picture:{}", img_path)
            continue
        height, width = img.shape[0:2]
        cropped = img[0:height-40, 0:width]
        adjusted_img_abs_path = adjusted_dir + '/' + img_path
        cv2.imwrite(adjusted_img_abs_path, cropped)
        cnt += 1

写在最后

写这个程序主要还是为了学习html解析和锤炼一下python编程，虽然写完了之后回过头来看确实没啥值得称道的地方，就把代码放这里供大家一起参考了：

https://gitee.com/tomocat/zhi-hu-picture-crawler

另外此程序的主要目的仅仅是将我搜集图片和剔除水印的过程自动化而已，还是再告诫大家一下不要因为爬虫给别人的服务器带来压力。

Reference

[1] https://www.cnblogs.com/jxlsblog/p/10445066.html