Recognize result difference in "view debug mode" and in "API"

Environment

• Tesseract Version:
  tesseract 4.0.0-beta.1-232-g45a6
  leptonica-1.74.4
  libjpeg 8d : libpng 1.2.50 : libtiff 4.0.8 : zlib 1.2.8

• Platform:
  Linux debian 4.9.0-6-amd64 #1 SMP Debian 4.9.88-1+deb9u1 (2018-05-07) x86_64 GNU/Linux

Current Behavior:

The result in view debug mode difference of my program that used libtesseract.

I run view debug mode use: tesseract input.jpg output -l chi_tra --psm 5 segmode inter

I was trained the chi_tra.traineddata my self, about used 46 types chinese fonts, the training text is chi_tra.training_text in langdata project.

even use tessdata_best/chi_tra.traineddata, the result also different.

For example:

• input image
  

• View debug mode output

tin this mode, i select a blob from image, the result is correct. ( "事" is the correct char)

• part of source code in my program

    api->SetPageSegMode(tesseract::PageSegMode::PSM_SINGLE_CHAR);
    // i corped a single char block from input image.
    api->SetImage(pix_char);
    api->SetSourceResolution(300);
    api->SetRectangle(0, 0, w, h);
  
    string ocr_result;
    if (api->Recognize(0) == 0) {
        ocr_result = (std::string) api->GetUTF8Text();
        cout << ocr_result;
    }
  

in my own program, the result is "蟬", not only this char, most of char in my program was wrong

But, most of char in view debug mode is correct!!

So, I Dont know that's my fault or tesseract's bug.

Reproduce the discrepancy with the supplied image, chi_tra.traineddata, and the tesseract ... --psm 5 segmode inter command. Compare that path with the program's SetPageSegMode(PSM_SINGLE_CHAR), SetImage, SetSourceResolution, SetRectangle, and Recognize calls. Done means identifying why the outputs differ and fixing or documenting the cause.