Python结合Tesseract-OCR训练库识别汉字的过程详解

一、安装python

例如，安装路径为：C:\rtkapp\python-3.8.0

二、安装opencv

三、安装tesseract-ocr

 安装完成后，在系统环境变量path中，添加安装路径C:\rtkapp\Tesseract-OCR

四、打开python安装pytesseract

 五、安装java运行环境

例如，安装版本为：jdk-8u191-windows-x64

 添加和配置系统环境变量

六、安装jTessBoxEditor

例如，安装版本为：jTessBoxEditor1.5

 七、打开jTesseBoxEditor为编辑汉字设置字体为宋体

八、用jTesBoxEditor合并选择的图片文件sample1.png

 保存结果如图：

 九、用命令生成box文件

例如，图片文件为：fmc.font.exp1.tif

运行命令为：tesseract.exe -l chi_sim fmc.font.exp1.tif fmc.font.exp1 batch.nochop makebox

生成的box文件为：fmc.font.exp1.box

十、浏览打开图片文件fmc.font.exp1.tif

用Box View调整X、Y、W、H使汉字正好在方框内，如汉字识别错误，在Char栏修改即可，

全部完成后，按Save保存即可，如下图：

十一、运行下面python脚本生成自定字库fmc.traineddata

import os
import cv2
import time
import pytesseract
import numpy as np
import subprocess
from PIL import Image

#Change work path
workpath="C:\\picdata";
curpath=os.getcwd();
os.chdir(workpath);

#Create a default font properties file
strcmd="echo normal 0 0 0 0 0>font_properties";
print(workpath+">"+strcmd);
print(subprocess.getoutput(strcmd)+"\n");

#Run tesseract for training
strcmd="tesseract.exe -l chi_sim fmc.font.exp1.tif fmc.font.exp1 nobatch box.train";
print(workpath+">"+strcmd);
print(subprocess.getoutput(strcmd)+"\n");

#Compute the character set
strcmd="unicharset_extractor.exe fmc.font.exp1.box";
print(workpath+">"+strcmd);
print(subprocess.getoutput(strcmd)+"\n");
strcmd="mftraining -F font_properties -U unicharset -O fmc.unicharset fmc.font.exp1.tr ";
print(workpath+">"+strcmd);
print(subprocess.getoutput(strcmd)+"\n");

#Run clustering
strcmd="cntraining.exe fmc.font.exp1.tr";
print(workpath+">"+strcmd);
print(subprocess.getoutput(strcmd)+"\n");

#Rename files
strcmd="move normproto fmc.normproto";
print(workpath+">"+strcmd);
print(subprocess.getoutput(strcmd)+"\n");
strcmd="move inttemp fmc.inttemp";
print(workpath+">"+strcmd);
print(subprocess.getoutput(strcmd)+"\n");
strcmd="move pffmtable fmc.pffmtable";
print(workpath+">"+strcmd);
print(subprocess.getoutput(strcmd)+"\n");
strcmd="move shapetable fmc.shapetable";
print(workpath+">"+strcmd);
print(subprocess.getoutput(strcmd)+"\n");

#Create tessdata
strcmd="combine_tessdata.exe fmc";
print(workpath+">"+strcmd);
print(subprocess.getoutput(strcmd)+"\n");

tessdata_path="";
strpath=os.getenv("path").split(";");
for kkk in strpath:
    if kkk.lower().find("tesseract")>0:
        tessdata_path=kkk+"\\tessdata";
        break;

#Copy tessdata
print(tessdata_path);
#Create Tessdata
if len(tessdata_path)>0:
    strcmd="copy fmc.traineddata "+tessdata_path;
    print(workpath+">"+strcmd);
    print(subprocess.getoutput(strcmd)+"\n");

十二、用生成的自定义字库fmc识别图片汉字

import os
import cv2
import pytesseract

pytesseract.pytesseract.tesseract_cmd = 'tesseract.exe'
img = cv2.imread('C:\\picdata\\sample1.png')
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

print(pytesseract.image_to_string(img,lang="fmc"))

运行结果如下图：