Regression in extracting text from Excel TIF image

[jwmepiq]

Environment

ExcelTest_Bug.zip
ExcelTest_text_TesseractV4.txt

Tesseract Version: 5.2.0 vs. 4.1.1.-rc2-37-gcla5
Ubuntu 20.04.3 LTS

Current Behavior:

With the attached TIF image of an Excel file (in the zip), Tesseract version 5.2.0 extracts a minimal amount of text (only a single line "hiding rows 15 through 20"). However, in prior versions of Tesseract, namely the version 4.1.1 version noted above, but likely other versions as well, the amount of text extracted from the same TIF image is significantly larger (multiple lines of text, approximately 1K of text over multiple pages). Attached a separate text file with the output of the V4.x text output.

Expected Behavior:

Expecting version 5.2+ of Tesseract to at least replicate the behavior of prior versions in extracting text from this sample TIF.

Suggested Fix:

Correct the text extraction to match the output from previous Tesseract versions. Concerned with Tesseract's regression in ability to extract text from Excel files.

[vamsiyadavmolli]

This is an issue in 5.3 version as well. Tried using both best/fast trained data still we see this issue.

[dhairyagupta2603]

Hey, Just wanted to know the status of this issue. Can I take this up?

[stweil]

I can confirm the issue with the latest code. What about releases between 4.1.1 (working) and 5.2.0 (not working)? Can we narrow down which release introduced the regression?

[stweil]

According to git bisect the regression was introduced by commit 842cca1. So release5.0.0-beta-20210916 was the last one without this issue.

[stweil]

Pull request #4136 fixes the regression.

[stweil]

@jwmepiq, thank you for reporting this nasty regression.

[tfmorris]

Now that #4136 has been merged and 5.3.3 released, I assume this can be closed

[stweil]

Yes, thanks.