Tesseract fails to recognize big text

[Euclisa]

Environment

• Tesseract Version: 4.1.1
• Platform: Linux arch 5.10.46-1-lts

Current Behavior:

Tesseract can't recognize text on this image.

Suggested Fix:

I reduced size of the image and it works, but shouldn't Tesseract recognize it without any extra transformations?

[nocun]

I think this may be the issue with the text being white-on-black. Can you check it by manually inverting colors and trying again?

[wollmers]

Tried it with with tesseract version 5 and it works perfect:

$ tesseract big.jpg big.jpg  --tessdata-dir /usr/local/share/tessdata txt hocr
Tesseract Open Source OCR Engine v5.0.0-alpha-773-gd33ed with Leptonica
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 1841
$ cat big.jpg.txt
Pot: 192,100

[nocun]

@wollmers Did you use tessdata best or fast? I tried Tesseract 5 with fast and it failed.

[Euclisa]

@nocun Yes, I tried and it works. But I have some pictures with the same text size and on some of them Tesseract fails.

[wollmers]

@wollmers Did you use tessdata best or fast? I tried Tesseract 5 with fast and it failed.

None of best or fast.

Just downloaded today for a fresh compile/installation on another server

wget https://github.com/tesseract-ocr/tessdata/raw/master/eng.traineddata

$ tesseract big.jpg big.w4.oem1.jpg  --oem 1 --tessdata-dir /usr/local/share/tessdata txt hocr
Tesseract Open Source OCR Engine v5.0.0-alpha-20210401 with Leptonica
Estimating resolution as 1841
$ cat  big.w4.oem1.jpg.txt
Pot: 192,100

[stweil]

Here is the image used for this issue (the link above might not be available in some weeks).

[wollmers]

Here is the image used for this issue (the link above might not be available in some weeks).

Images that small (33 KB) should be in the issue thread.

If I test an issue, I try to record the case in an own github repository like this: https://github.com/wollmers/ocr-tess-issues/tree/main/issues/issue_3304_big_text.

[amitdo]

Yes, I tried and it works.

So, no issue recognizing the image you provided. Getting different results with different models is normal.

But I have some pictures with the same text size and on some of them Tesseract fails.

So provide 2-3 examples. Use the latest 5.0.0 codebase.

[amitdo]

No feedback from the OP.