Background.

AOI software needs to use the OCR feature to recognize the texts on the chips. Because our vision software is based on OpenCV, so the first choice is text module in opencv_contrib.

Procedures.

1. OCR module is not in standard OpenCV package. It is in text module of OpenCV_Contrib. It can be downloaded from opencv_contrib.

2. The core of OCR is using Tesseract, and Tesseract depends on Leptonica, so need to build Leptonica and Tesseract first.

3. Get the Leptonica from https://github.com/charlesw/tesseract-vs2012. This solution can directly build. The output is liblept171d.dll and liblept171d.lib.

4. Get the Tesseract from https://github.com/tesseract-ocr/tesseract. Create the .\include\leptonica folder, copy all the header file from .\tesseract-vs2012\liblept\include (The root folder is in step 3) to it. Create the .\lib folder, copy all the  step 3 build generated files(The .dll and .lib) in .\tesseract-vs2012\build\lib to it.  Set the project property of tesseract, change the include folder path "..\..\..\include" and "..\..\..\include\leptonica" to "..\..\include" and "..\..\include\leptonica". Then can build the Tessrect project, the output is libtesseract304d.dll and libtesseract304d.lib.

The AddIdTohOCR has compile error, change the content as below can solve the problem.

static void AddIdTohOCR(STRING* hocr_str, const std::string base, int num1, int num2) {

    unsigned long bufsize = base.length() +  * kMaxIntSize;

    char *id_buffer = new char[bufsize];

    if (num2 >= ) {

        snprintf(id_buffer, bufsize - , "%s_%d_%d", base.c_str(), num1, num2);

    }

    else {

        snprintf(id_buffer, bufsize - , "%s_%d", base.c_str(), num1);

    }

    id_buffer[bufsize - ] = '\0';

    *hocr_str += " id='";

    *hocr_str += id_buffer;

    *hocr_str += "'";

    delete[]id_buffer;

}

5. Use CMake to config the OpenCV solution. Copy the text module from opencv_contrib to .\OpenCV\sources\modules. Run Cmake_Gui, there are 3 options need to set. Lept_library, Tesseract_Include_Dir, Tesseract_Library. Tesseract_Include_Dir set to ../.../tesseract/API. After set, can run CMake to config and generate the solution.

6. Open the OpenCV solution. Open the .\modules\text\src\precomp.hpp file, change the include path to as below.

#ifdef HAVE_TESSERACT

//#include <tesseract/baseapi.h>

//#include <tesseract/resultiterator.h>

#include "baseapi.h"

#include "resultiterator.h"

#endif

If there are header files can not find errors, find and copy them from tesseract to the tesseract/API folder. As i test, below files need to be copied.

#include ""resultiterator.h""

#include "platform.h"

#include "apitypes.h"

#include "thresholder.h"

#include "unichar.h"

#include "tesscallback.h"

#include "publictypes.h"

#include "pageiterator.h"

#include "resultiterator.h"

#include "host.h"

#include "ltrresultiterator.h"

There may be a compile error in function static double NFA(int n, int k, double p, double logNT) with std::numeric_limits<double>::min(); Add below code before the function to fix it.

#undef max

#undef min

7. Download the language test data from https://github.com/tesseract-ocr/tessdata. What i use is the eng.traineddata. Put it to .\tesseract\tessdata.

8. After build OpenCV successfully, then you can create the TestOpenCV project with the below function, before running it, need to copy the liblept171d.dll and libtesseract304d.dll to the output folder(where the exe file is put).

using OCRTesseract =  cv::text::OCRTesseract;

void TestOCR()

{

    cv::Mat mat = cv::imread(".\\data\\OCRTest.png");

    if ( mat.empty() )

        return;

    std::string output_text;

    char *dataPath = "C:/tesseract-build/tesseract/tessdata";

    cv::Ptr<OCRTesseract> ptrOcr = OCRTesseract::create(dataPath);

    ptrOcr->run(mat, output_text );

    cout << output_text << endl;

}

9. The Tesseract is sensitive to the text orientation. So need to make the text face up to be better recognized.