python利用pytesseract 实现本地识别图片文字
(编辑:jimmy 日期: 2024/11/2 浏览:3 次 )
#!/usr/bin/env python3 # -*- coding: utf-8 -*- import glob from os import path import os import pytesseract from PIL import Image from queue import Queue import threading import datetime import cv2 def convertimg(picfile, outdir): '''调整图片大小,对于过大的图片进行压缩 picfile: 图片路径 outdir: 图片输出路径 ''' img = Image.open(picfile) width, height = img.size while (width * height > 4000000): # 该数值压缩后的图片大约 两百多k width = width // 2 height = height // 2 new_img = img.resize((width, height), Image.BILINEAR) new_img.save(path.join(outdir, os.path.basename(picfile))) def baiduOCR(ts_queue): while not ts_queue.empty(): picfile = ts_queue.get() filename = path.basename(picfile) outfile = 'D:\Study\pythonProject\scrapy\IpProxy\port_zidian.txt' img = cv2.imread(picfile, cv2.IMREAD_COLOR) print("正在识别图片:\t" + filename) message = pytesseract.image_to_string(img,lang = 'eng') message = message.replace('', '') message = message.replace('\n', '') # message = client.basicAccurate(img) # 通用文字高精度识别,每天 800 次免费 #print("识别成功!")) try: filename1 = filename.split('.')[0] filename1 = ''.join(filename1) with open(outfile, 'a+') as fo: fo.writelines('\'' + filename1 + '\'' + ':' + message + ',') fo.writelines('\n') # fo.writelines("+" * 60 + '\n') # fo.writelines("识别图片:\t" + filename + "\n" * 2) # fo.writelines("文本内容:\n") # # 输出文本内容 # for text in message.get('words_result'): # fo.writelines(text.get('words') + '\n') # fo.writelines('\n' * 2) os.remove(filename) print("识别成功!") except: print('识别失败') print("文本导出成功!") print() def duqu_tupian(dir): ts_queue = Queue(10000) outdir = dir # if path.exists(outfile): # os.remove(outfile) if not path.exists(outdir): os.mkdir(outdir) print("压缩过大的图片...") # 首先对过大的图片进行压缩,以提高识别速度,将压缩的图片保存与临时文件夹中 try: for picfile in glob.glob(r"D:\Study\pythonProject\scrapy\IpProxy\tmp\*"): convertimg(picfile, outdir) print("图片识别...") for picfile in glob.glob("tmp1/*"): ts_queue.put(picfile) #baiduOCR(picfile, outfile) #os.remove(picfile) print('图片文本提取结束!文本输出结果位于文件中。' ) #os.removedirs(outdir) return ts_queue except: print('失败') if __name__ == "__main__": start = datetime.datetime.now().replace(microsecond=0) t = 'tmp1' s = duqu_tupian(t) threads = [] try: for i in range(100): t = threading.Thread(target=baiduOCR, name='th-' + str(i), kwargs={'ts_queue': s}) threads.append(t) for t in threads: t.start() for t in threads: t.join() end = datetime.datetime.now().replace(microsecond=0) print('删除耗时:' + str(end - start)) except: print('识别失败')
实测速度慢,但用了多线程明显提高了速度,但准确度稍低,同样高清图片,90百分识别率。还时不时出现乱码文字,乱空格,这里展现不了,自己实践吧,重点免费的,随便识别,通向100张图片,用时快6分钟了,速度慢了一倍,但是是免费的,挺不错的了。
以上就是python利用pytesseract 实现本地识别图片文字的详细内容,更多关于python 识别图片文字的资料请关注其它相关文章!
下一篇:详解Python中@staticmethod和@classmethod区别及使用示例代码