码迷,mamicode.com
首页 > 其他好文 > 详细

使用 selenium 实现谷歌以图搜图爬虫

时间:2020-04-23 16:18:35      阅读:75      评论:0      收藏:0      [点我收藏+]

标签:expec   copyfile   get   bmp   xtend   谷歌   join   make   gif   

实现思路

原理非常简单,就是利用selenium去操作浏览器,获取到想要的链接,然后进行图片的下载,和一般的爬虫无异。

用到的技术:multiprocessing,selenium,xpath,requests

以下按照代码执行的顺序进行讲解。

首先导入需要的包

# coding=utf-8
import base64
import hashlib
import os
import re
import shutil
import time
from multiprocessing import Pool, cpu_count

import requests
import tqdm
from colorama import Fore
from selenium import webdriver
from selenium.common.exceptions import (ElementNotVisibleException,
                                        StaleElementReferenceException)
from selenium.webdriver.common.by import By
from selenium.webdriver.common.keys import Keys
from selenium.webdriver.support import expected_conditions as EC
from selenium.webdriver.support.wait import WebDriverWait

定义一个 run()函数,作为入口。这里使用多进程技术,同时打开多个浏览器进行图片爬取。

def run():
    num_process = cpu_count() # 进程数设置为cpu核心数
    pool = Pool(num_process) # 建立一个进程池
    filelist = []
    upload = r"./upload" # 需要进行上传的图片文件夹
    getfilelist(upload, filelist)  # 递归查找文件夹里面所有的图片文件
    result = partition(filelist, num_process) # 将图片文件列表平均分为几个list,每个进程跑一部分
    pool.map_async(download_task, result) # 下载任务丢进进程池
    pool.close() # 不再允许加入进程池
    pool.join() # 等待进程完成

其中 getfilelist()函数是递归查找,工作中用得很多了。

EXTEND = [".bmp", ".jpg", ".jpeg", ".tif", ".tiff",
          ".jfif", ".png", ".gif", ".iff", ".ilbm"]
          
def is_img(img_path):
    # 根据后缀判断是否为图片
    ext = os.path.splitext(img_path)[1]
    if ext in EXTEND:
        return True
    else:
        return False
        
def getfilelist(path, filelist):
    file = os.listdir(path)
    for im_name in file:
        if os.path.isdir(os.path.join(path, im_name)):
            getfilelist(os.path.join(path, im_name), filelist)
        else:
            if is_img(im_name):
                name = os.path.join(path, im_name)
                filelist.append(name)

partition()函数用于将一个列表均分为几份,以便实现多进程。

def partition(ls, size):
    num_per_list = len(ls)//size
    result = []
    if num_per_list*size == len(ls):
        for i in range(size):
            result.append(ls[num_per_list*i:num_per_list*(i+1)])
    else:
        for i in range(size-1):
            result.append(ls[num_per_list*i:num_per_list*(i+1)])
        result.append(ls[num_per_list*(size-1):])
    return result

download_task()为具体的下载任务,一个task实例化一个GoogleSearcher类,遍历自己的图片列表进行以图搜图。

def download_task(filelist):
    searcher = GoogleSearcher(
        download=r"./download")
    for file in filelist:
        searcher.simple_file_run(file)  # 上传单张图并进行以图搜图

GoogleSearcher类比较长,在注释中进行讲解。

USERNAME = os.environ[‘USERNAME‘]
class GoogleSearcher:
    def __init__(self, download="download", sleep_time=1):
        super().__init__()
        self._download = download # 下载文件夹
        self.sleep_time = sleep_time  # 下载页面时等待时间
        self.header = {
            "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/80.0.3987.163 Safari/537.36"}

        os.makedirs(self._download, exist_ok=True)  # 创建下载文件夹

        self.option = webdriver.ChromeOptions()
        # self.option.add_argument("--user-data-dir=" + f"C:/Users/{USERNAME}/AppData/Local/Google/Chrome/User Data/")
        # self.option.add_argument("headless")  # if use headless, may failed.
        self.option.add_argument("disable-gpu")
        self.driver = webdriver.Chrome(options=self.option) # 以上为浏览器对象创建

    def upload_img_get_html(self, file):
    	# 上传图片并转到图片列表页面
        print(
            f"{Fore.GREEN} Begin to upload image {os.path.split(file)[1]} {Fore.RESET}")
        self.driver.get("https://www.google.com/imghp")

        # 等待相机按钮出现
        condition_1 = EC.visibility_of_element_located(
            (By.CLASS_NAME, "LM8x9c"))
        WebDriverWait(self.driver, timeout=20,
                      poll_frequency=0.5).until(condition_1)
        # 相机按钮出现后点击
        image_button = self.driver.find_element_by_class_name("LM8x9c")
        image_button.send_keys(Keys.ENTER)

        # 等待出现上传图片字样
        condition_2 = EC.visibility_of_element_located(
            (By.ID, "dRSWfb"))
        WebDriverWait(self.driver, timeout=20, poll_frequency=0.5).until(
            condition_2)

        # 点击上传图片
        upload = self.driver.find_element_by_xpath(‘//*[@id="dRSWfb"]/div/a‘)
        upload.send_keys(Keys.ENTER)

        # 找到上传图片的控件
        condition_3 = EC.visibility_of_element_located(
            (By.ID, ‘awyMjb‘))
        WebDriverWait(self.driver, timeout=10, poll_frequency=0.5).until(
            condition_3)
        input_ = self.driver.find_element_by_id(‘awyMjb‘)

        # 因为上传图片的控件是一个input,直接将文件send就行
        input_.send_keys(file)
        print(f"{Fore.GREEN} uploaded {Fore.RESET}")

        # 页面转向另一页
        condition_4 = EC.visibility_of_element_located(
            (By.XPATH, ‘//*[@id="top_nav"]‘))
        WebDriverWait(self.driver, timeout=20,
                      poll_frequency=0.5).until(condition_4)
        # 等待片刻
        time.sleep(self.sleep_time)

        # print(driver.current_url)
        # print(driver.page_source)
        print(f"{Fore.GREEN} Finish download source code{Fore.RESET}")
        return self.driver.page_source

    def highlight(self, element):
        self.driver.execute_script(
            "arguments[0].setAttribute(‘style‘, arguments[1]);", element, "background: yellow; border: 2px solid red;")

    def wait_and_click(self, xpath):
        #  Sometimes click fails unreasonably. So tries to click at all cost.
        try:
            w = WebDriverWait(self.driver, 15)
            elem = w.until(EC.element_to_be_clickable((By.XPATH, xpath)))
            elem.click()
            self.highlight(elem)
        except Exception as e:
            print(‘Click time out - {}‘.format(xpath))
            print(‘Refreshing browser...‘)
            self.browser.refresh()
            time.sleep(2)
            return self.wait_and_click(xpath)
        return elem

    def get_extension_from_link(self, link, default=‘jpg‘):
    # 获取文件后缀
        splits = str(link).split(‘.‘)
        if len(splits) == 0:
            return default
        ext = splits[-1].lower()
        if ext == ‘jpg‘ or ext == ‘jpeg‘:
            return ‘jpg‘
        elif ext == ‘gif‘:
            return ‘gif‘
        elif ext == ‘png‘:
            return ‘png‘
        else:
            return default

    def base64_to_object(self, src):
    # base64 解码
        header, encoded = str(src).split(‘,‘, 1)
        data = base64.decodebytes(bytes(encoded, encoding=‘utf-8‘))
        return data

    def download_images(self, links, download_dir):
    # 下载图片
        total = len(links)
        for index, link in enumerate(links):
            try:
                if len(link) < 100:
                    print(‘Downloading {} : {} / {}‘.format(link, index + 1, total))
                else:
                    print(
                        ‘Downloading {} : {} / {}‘.format(link[:100], index + 1, total))
                        # 链接过长,只打印部分
                if str(link).startswith(‘data:image/jpeg;base64‘):
                # base64编码的jpg图片
                    response = self.base64_to_object(src=link)
                    ext = ‘jpg‘
                    is_base64 = True
                elif str(link).startswith(‘data:image/png;base64‘):
                # base64编码的png图片
                    response = self.base64_to_object(src=link)
                    ext = ‘png‘
                    is_base64 = True
                else:
                # 图片超链接
                    response = requests.get(link, stream=True, timeout=5)
                    ext = self.get_extension_from_link(link=link)
                    is_base64 = False

                path = os.path.join(download_dir, str(index).zfill(4)+"."+ext)
                try:
                    with open(path, "wb") as f:
                    # base64图片和超链接图片两种保存方法
                        if is_base64:
                            f.write(response)
                        else:
                            shutil.copyfileobj(response.raw, f)
                except Exception as e:
                    print(‘Save failed - {}‘.format(e))

                del response
            except Exception as e:
                print(‘Download failed - ‘, e)
                continue

    def get_full_resolution_links(self):
        print(‘[Full Resolution Mode]‘)
        time.sleep(1)
        elem = self.driver.find_element_by_tag_name("body")
        print(‘Scraping links‘)
        self.wait_and_click(‘//div[@data-ri="0"]‘)
        time.sleep(1)
        links = []
        count = 1
        last_scroll = 0
        scroll_patience = 0
        while True:
            try:
                xpath = ‘//div[@id="islsp"]//div[@class="v4dQwb"]‘
                div_box = self.driver.find_element(By.XPATH, xpath)
                self.highlight(div_box)
                xpath = ‘//img[@class="n3VNCb"]‘
                img = div_box.find_element(By.XPATH, xpath)
                self.highlight(img)
                xpath = ‘//div[@class="k7O2sd"]‘
                loading_bar = div_box.find_element(By.XPATH, xpath)
                # 等待图片加载,如果加载不完,获取到的是 base64 编码的图片
                while str(loading_bar.get_attribute(‘style‘)) != ‘display: none;‘:
                    time.sleep(0.1)
                src = img.get_attribute(‘src‘)
                if src is not None:
                    links.append(src)
                    if len(src) < 100:
                        print(‘%d: %s‘ % (count, src))
                    else:
                        print(‘%d: %s‘ % (count, src[:100])) # 如果太长,只打印一部分
                    count += 1
            except StaleElementReferenceException:
                pass
            except Exception as e:
                print(
                    ‘[Exception occurred while collecting links from google_full] {}‘.format(e))
            scroll = self.driver.execute_script("return window.pageYOffset;") # 页面滚动的位置
            if scroll == last_scroll:
            # 页面滚动1
                scroll_patience += 1
            else:
                scroll_patience = 0
                last_scroll = scroll
            if scroll_patience >= 30:
            #页面滚动30,停止
                break
            elem.send_keys(Keys.RIGHT)
        links = list(dict.fromkeys(links)) # 链接去重
        print(‘Collect links done. Total: {}‘.format(len(links)))
        return links

    def simple_file_run(self, img):
        # 上传图片并进行搜索
        img_name = os.path.splitext(os.path.split(img)[1])[0] # 图片名
        parent_name = os.path.split(os.path.split(img)[0])[-1] # 图片的父级名字,用来区分图片的类别
        print("--> Processing image:  {}  ".format(img_name))
        download_dir = os.path.join(self._download, parent_name, img_name)
        os.makedirs(download_dir, exist_ok=True)  
        html_source = self.upload_img_get_html(img)  # 上传图片,到搜索结果页
        similar_img_href = self.driver.find_element_by_xpath(
            ‘//div[@class="e2BEnf U7izfe"]/h3/a‘)
        similar_img_href.click()  # 查找“类似图片”的链接并点击,进入图片列表页
        links = self.get_full_resolution_links()  # 将所有图片的大图链接进行收集
        self.download_images(links, download_dir)  # 下载这些大图
        print("{}Image {} finished\n{}".format(
            Fore.GREEN, img_name, Fore.RESET))

整个流程就跟打开浏览器进行操作一样,难点在于如何控制速度,不被谷歌反爬,不然出现谷歌验证码,破解是不可能的,就要帮它免费打码了。

有何用途

当你需要训练一个图片分类的模型,手头上图片有限,那就可以用这个方法,每一张图都找跟它相似的,轻轻松松就把训练集扩大了几十倍(理想情况,不被反爬的话)。

参考

  1. https://github.com/YoongiKim/AutoCrawler
  2. https://github.com/Cyberist-Edgar/Google_Image_Searcher

使用 selenium 实现谷歌以图搜图爬虫

标签:expec   copyfile   get   bmp   xtend   谷歌   join   make   gif   

原文地址:https://www.cnblogs.com/willwell/p/google_image_search.html

(0)
(0)
   
举报
评论 一句话评论(0
登录后才能评论!
© 2014 mamicode.com 版权所有  联系我们:gaon5@hotmail.com
迷上了代码!