Tesseract OCR: Parameter for Font Size (Single Character)

Question

I want to use Tesseract to recognize a single noiseless character with a typical font (ex. Times New Roman, Arial, etc. No weird font). The input image just contains the character, so the input image size is equivalent to the font size.

I already set the page segmentation mode to single character, but the result is still not satisfactory, with error rate of ~50%.

I thought that I can improve my result if I tell Tesseract what my font size will be. Is there such parameter? Also, if it exists, does python-tesseract (Python wrapper) allow to tweak this parameter?

Wait a moment—you have a clear image ("noiseless") of a single character ("no weird font"), and not even that is recognized correctly? You aren't trying to distinguish `|` from `l` and from `1` and from `I`, aren't you? Or '0' from 'O' and `×` from `x`? This sounds really strange. — Alfe, Jan 23 '14 at 09:32
It is not as challenging as distinguishing 1 from l. For instance, I get M for B. The size of the image is very small though (~10 by 10 pixels), so that is where I think the problem is arising from... — Min Joon Seo, Jan 23 '14 at 21:07
Oh, 10x10, yes, that's the reason. No chance of increasing that one? — Alfe, Jan 24 '14 at 08:07
Maybe you can post some example input images in your question (images can be included in the question) so we can see what "clearly recognizable images" aren't recognized by Tesseract? You could also maybe improve recognition by applying some image processing before. I imagine blowing up the image by a factor of 5x5 and smoothing edges before running the OCR. But I would not expect too much. — Alfe, Jan 24 '14 at 08:17

score 3 · Answer 1 · answered Oct 20 '14 at 10:16

If your font size is too small then increase the image height and width ,so that tesseract will provide more accurate output.

        var srcImage1 = System.Drawing.Image.FromFile(@"D:\Image\font_english.jpg");
        var newWidth1 = (int)(srcImage1.Width * 2);
        var newHeight1 = (int)(srcImage1.Height * 2);

        var image  = new Bitmap(srcImage1, new Size(newWidth1, newHeight1));
        var ocr = new Tesseract();

          ocr.Init(@"D:\OCRTEST\tessdata\", "eng", false);
          ocr.SetVariable("tessedit_char_whitelist", "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ-0123456789'?.;=,()");
          var result = ocr.DoOCR(image, Rectangle.Empty);
          foreach (Word word in result)
          {
              Response.Write(word.Text+" ");

          }

Tesseract OCR: Parameter for Font Size (Single Character)

1 Answers1