OCR识别表格中的参数及参数值

一、需求

识别固定表格中的文字，输出表格中的参数字典

二、整体思路

1. 找到一张含有表格的图片，利用mac自带的预览工具分析出图片每个单元格的对角线坐标，
2. 使用程序根据图片文件和对角线坐标数组，分割图片为很多个小图片，
3. 采用tesseract库识别单个单元格中的文字，并将识别出来的所有文字输出到同一个文本文件中。

三、开发环境准备

Mac上安装VSCode，VSCode中安装python扩展，使用venv这个python自带的虚拟环境，建立测试venv工作区后利用.venv/bin/pip3安装tesseract和pillow类库

.venv/bin/pip3 install pillow
.venv/bin/pip3 install tesseract

.venv/bin/pip3 install pytesseract
.venv/bin/pip3 install opencv-python
.venv/bin/pip3 install numpy
.venv/bin/pip3 install scikit-image

 

四、开发步骤

建立根目录及相关功能目录，并使用VSCode打开根目录

 编写图片分割程序

import os
from PIL import Image

img = "imgs2txt/imgs/合格证.jpg"
pix_a = [(20,3,428,20),     (428,3,700,20),
         (20,20+1,428,20*2),(428,20+1,700,20*2)]

def cut_by_pix_a(img, xys):
    number_of_pic = len(xys)
    image = Image.open(img)
    print(f'这个图像是{image.width} x {image.height}')
    name1, name2 = os.path.dirname(img) + "/splits/" ,os.path.basename(img).split(".")[-2]+"_cut_"
    for p in xys:
        name3 = name1+name2+str(p)+".jpg"
        im2 = image.crop(p)
        im2.save(name3)
    print(f"一共完成了{len(xys)}个文件.")

cut_by_pix_a(img, pix_a)

编写图片增强处理器

分割后发现图片非常模糊识别时，汉子无法识别出来，所以编写图片锐化程序，提高图片识别精度

import cv2
import numpy as np
from skimage import filtersdef preprocess_image_old(img_path):img = cv2.imread(img_path)# 1.灰度化gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)# 2.去操（中值滤波）denoised = cv2.medianBlur(gray, 3)# 3.自适应二值化binary = cv2.adaptiveThreshold(denoised, 255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY, 11, 2)# 4.锐化处理（拉普拉斯算子）kernel = np.array([[0, -1, 0], [-1, 5,-1], [0, -1, 0]])sharpened = cv2.filter2D(binary, -1, kernel)# 5.形态学操作（去除小噪点）kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (2,2))cleaned = cv2.morphologyEx(sharpened, cv2.MORPH_OPEN, kernel)return cleaneddef preprocess_image(img_path):img = cv2.imread(img_path)# 1.灰度化# gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)# 1.去操（中值滤波）denoised = cv2.medianBlur(img, 3)# 2.自适应二值化binary = cv2.adaptiveThreshold(denoised, 255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY, 11, 2)# 3.锐化处理（拉普拉斯算子）kernel = np.array([[0, -1, 0], [-1, 5,-1], [0, -1, 0]])sharpened = cv2.filter2D(binary, -1, kernel)# 5.形态学操作（去除小噪点）kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (2,2))cleaned = cv2.morphologyEx(sharpened, cv2.MORPH_OPEN, kernel)return cleanedimg1_path = 'imgs2txt/imgs/splits/合格证_cut_(20, 3, 106, 20).jpg'
r = preprocess_image(img1_path)
cv2.imwrite('imgs2txt/imgs/cleaned/合格证_cut_(20, 3, 106, 20).jpg', r)
print(r)

 

 

编写单个图片识别程序

 

import pytesseract
from PIL import Imageimg1_path = 'imgs2txt/imgs/splits/合格证_cut_(20, 3, 106, 20).jpg'def basic_ocr(img_path):img = Image.open(img_path)text = pytesseract.image_to_string(img, lang='chi_sim')return textdef advanced_ocr(img_path):custom_config = r'--oem 3 --psm 6 -c tessedit_char_whitelist=0123456789'text = pytesseract.image_to_string(Image.open(img_path),lang='chi_sim',config=custom_config)return textprint("基本识别结果" + basic_ocr(img1_path))
print("高级识别结果" + advanced_ocr(img1_path))

 

待处理的图片

 

......未完待续