python easyocr和cv2实现名片识别及裁剪摆正
公司希望能做一个名片裁剪和识别功能。我来开发小程序,在寻找合适的api的途中,因嫌弃乙方弄得太慢,自己花了半天百度搞了个python版本的。很久没用python了好多基础方法都忘记了,实现的很草率。
效果图,识别的准确度还得微调,资源包应该还得训练啥的。
命令:
pip install opencv-python==4.1.2.30 -i https://pypi.tuna.tsinghua.edu.cn/simple
python代码(与原作者代码有差异):
# -*- coding:utf-8 -*- import easyocr import time import cv2 import numpy as np from PIL import Image # 可信度最小阈值 BASE_MIN = 0.1 BASE_ROTA = 0 time1 = time.time() url = './1.jpg' reader = easyocr.Reader(['ch_sim','en'],gpu=True) result = reader.readtext(url,batch_size =100,rotation_info=[90,180,270],low_text=0.4,slope_ths=0.3,beamWidth =1,min_size=50) numLis = [0,0,0,0] for i in result: numLis[i[3]] = numLis[i[3]]+1 for i in range(len(numLis)): if(numLis[i]>numLis[BASE_ROTA]): BASE_ROTA = i for i in result: if(float(i[2])>BASE_MIN and i[3] == BASE_ROTA): print(i[1]) image = cv2.imread(url) # 灰度图 img=cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) # 二值化 ret , thresh = cv2.threshold(img, 230, 255, cv2.THRESH_BINARY_INV) contours, hierarchy = cv2.findContours(cv2.Canny(img,80,100),cv2.RETR_LIST,cv2.CHAIN_APPROX_NONE) dot=[] # 用来保存所有轮廓返回的坐标点。 for c in contours: # 找到边界坐标 min_list=[] # 保存单个轮廓的信息,x,y,w,h,area。 x,y 为起始点坐标 x, y, w, h = cv2.boundingRect(c) # 计算点集最外面的矩形边界 min_list.append(x) min_list.append(y) min_list.append(w) min_list.append(h) min_list.append(w*h) # 把轮廓面积也添加到 dot 中 dot.append(min_list) max_area=dot[0][4] for inlist in dot: area=inlist[4] if area >= max_area: x=inlist[0] y=inlist[1] w=inlist[2] h=inlist[3] max_area=area # 在原图上画出最大的矩形 cv2.rectangle(image, (x, y), (x + w , y + h ), (0, 255, 0), 1) cv2.imwrite('big.jpg',image) img = Image.open('big.jpg') region = img.crop((x, y, x + w, y + h)) if(BASE_ROTA == 1): region = region.transpose(Image.ROTATE_90) elif(BASE_ROTA == 2): region = region.transpose(Image.ROTATE_180) elif(BASE_ROTA == 3): region = region.transpose(Image.ROTATE_270) region.save('result.jpg') time2 = time.time() print("绘制结束,用时:",time2-time1)
需要修改一下easyocr的包返回一个识别的方向供图片裁剪摆正:
参考链接:
使用cv2报错:https://blog.csdn.net/iLOVEJohnny/article/details/126466515;
下载cv2:https://blog.csdn.net/songyuc/article/details/102855144
识别文字:https://blog.csdn.net/weixin_47046791/article/details/124149260
最大矩形以及裁剪:https://blog.csdn.net/aqqwvfbukn/article/details/120787480
肖cc
QQ2398506993