Strange behaviour of Tesseract

[Gilad Pellaeon]

Hi,

I am new to Tesseract. I searched for an OCR library, found Tesseract and now I want to use it for a specific measure protocol.

I built Tesseract 5.3.2 from source and the dependencies leptonica-1.83, libpng and OpenJPEG for Windows with the Latex Visual C++ compiler for Windows, x64.

Then I did some first tests based on the examples from the documentation ( Basic_example and SetRectangle_example). As data set I use eng.traineddata from the testdata_best repo.

Now, I have a behaviour which I can't classify. I tried to recognize a float value in a given rectangle (with SetRectangle ). Tesseract didn't converted it (empty return). Then I manually copied the rectangle and saved it in a new file (see attached Single_Number.png). Then I tried this file without the SetRectangle call. Now it works.

The attached Protocol_table.png is the original image, but I removed all other stuff in the picture. So it's empty except the number at the original position. Now I have the following behaviour: in DEBUG mode the conversion works, in RELEASE mode not.

I also tried to slighty enlarge the rectangle area (see last SetRectangle call in the code below). But now I got a runtime exception. The resolution of the picture is 2625x1682. So there should be no buffer overflow?!

Am I doning something wrong here? Or what's the problem for this behaviour?

This is my basic code:

//std includes
#include <iostream>

//tesseract includes
#include "tesseract/baseapi.h"

//Leptonica includes
#include "allheaders.h"



//!
int main()
{
    tesseract::TessBaseAPI api;
    // Initialize tesseract-ocr with English, without specifying tessdata path
    if (api.Init(nullptr, "eng"))
    {
        std::cout << "Could not initialize tesseract." << std::endl;
        return 1;
    }

    //
    Pix* image = pixRead("D:/projects/cpp/Tesseract-Test/Protocol_Table.png");
    //Pix* image = pixRead("D:/projects/cpp/Tesseract-Test/Single_Number.png");
    api.SetImage(image);
    // Restrict recognition to a sub-rectangle of the image
    // SetRectangle(left, top, width, height)
    api.SetRectangle(807, 1393, 93, 49);
    //api.SetRectangle(707, 1293, 193, 149);
    // Get OCR result
    char* outText = api.GetUTF8Text();
    if (outText)
        printf("OCR output:\n%s", outText);

    // Destroy used object and release memory
    api.End();    
}

[Zdenko Podobny]

https://github.com/tesseract-ocr/tesseract/issues/845

Zdenko

št 14. 9. 2023 o 16:49 Gilad Pellaeon <ld.pe...@gmail.com> napísal(a):

--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/626e06c6-ea15-45d2-86da-1bba6c069e1cn%40googlegroups.com.

[Gilad Pellaeon]

Is it still broken in version 5? The thread you posted is from 2017!

One thing I noticed in the meantime: I stored my PNGs with paint.net and auto bit depth recognition. The image is grayscale. So the image wasn't stored in 32bit.

Now, I forced to save the image 32bit color depth and now it works. So I assume it's a bug regarding the bit depth handling. This also explains the memory access violation althrough the pixel range doesn't violet the image pixel size. If the system internally assumes a bigger depth then it also wants to process bigger memory chunks.

Is this problem known? Otherwise I can create a bug report.

Best regards

[Zdenko Podobny]

Is it still broken in version 5? The thread you posted is from 2017!

Zdenko

št 14. 9. 2023 o 17:10 Gilad Pellaeon <ld.pe...@gmail.com> napísal(a):

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/d262537a-9065-4876-b435-071a8e596745n%40googlegroups.com.