sandwich pdf-rederer is not working. Getting gibberish text from output PDF.

[sanjay-nit]

Hi, Even after using --pdf-renderer=sandwich option; getting gibberish text when I select the text from output pdf and paste somewhere.

FYI: I'm using MacOS(M1) 15.0.1

Below are the steps I took.

1. pip install git+https://github.com/ocrmypdf/OCRmyPDF-EasyOCR.git
2. command: ocrmypdf --pdf-renderer sandwich image.pdf test.pdf

Version info:

easyocr==1.7.2
ocrmypdf==16.5.0
ocrmypdf-easyocr==0.2.1

Below are logs:

I'm attaching the PDF file I tested with.
image.pdf
ouput.pdf

I also tried processing image version of this PDF, this isn't working too.
command: ocrmypdf --pdf-renderer=sandwich --force-ocr --image-dpi 300 image.jpg test.pdf
image

[jbarlow83]

The text found in the PDF is precisely what EasyOCR detected in this case, so it must be struggling to make sense of the multiple colors and formatting.

From the debug log, the gibberish is coming directly from EasyOCR:

[2024-10-15 10:36:54,935] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline '|+_"' in-image bbox: 479, 2, 662, 2, 662, 56, 479, 56
[2024-10-15 10:36:54,936] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline '|+_"' PDF bbox: 479, 882, 662, 882, 662, 828, 479, 828
[2024-10-15 10:36:54,937] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline ']2' in-image bbox: 387, 17, 459, 17, 459, 61, 387, 61
[2024-10-15 10:36:54,937] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline ']2' PDF bbox: 387, 867, 459, 867, 459, 823, 387, 823
[2024-10-15 10:36:54,938] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline '@_@:' in-image bbox: 362, 27, 817, 27, 817, 144, 362, 144
[2024-10-15 10:36:54,938] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline '@_@:' PDF bbox: 362, 857, 817, 857, 817, 740, 362, 740

That's a fantastic test image. Please report the issue to EasyOCR to see if they will address it.

Regular OCRmyPDF with Tesseract seems to work okay-ish.

[sanjay-nit]

@jbarlow83 I'm not sure about this tool how it's using the easyocr under the hood. But below is what I got when I ran easyocr separately on this image. OCR is giving correct results on this image.

[sanjay-nit]

The text found in the PDF is precisely what EasyOCR detected in this case, so it must be struggling to make sense of the multiple colors and formatting.

From the debug log, the gibberish is coming directly from EasyOCR:

[2024-10-15 10:36:54,935] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline '|+_"' in-image bbox: 479, 2, 662, 2, 662, 56, 479, 56
[2024-10-15 10:36:54,936] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline '|+_"' PDF bbox: 479, 882, 662, 882, 662, 828, 479, 828
[2024-10-15 10:36:54,937] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline ']2' in-image bbox: 387, 17, 459, 17, 459, 61, 387, 61
[2024-10-15 10:36:54,937] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline ']2' PDF bbox: 387, 867, 459, 867, 459, 823, 387, 823
[2024-10-15 10:36:54,938] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline '@_@:' in-image bbox: 362, 27, 817, 27, 817, 144, 362, 144
[2024-10-15 10:36:54,938] - ocrmypdf_easyocr._pdf -   DEBUG -    1  Textline '@_@:' PDF bbox: 362, 857, 817, 857, 817, 740, 362, 740

That's a fantastic test image. Please report the issue to EasyOCR to see if they will address it.

Regular OCRmyPDF with Tesseract seems to work okay-ish.

@jbarlow83
I appreciate your quick closure of this GitHub issue! 😄

[alaminkouser]

Set language to English.

ocrmypdf.ocr(os.path.join(save_path, "input.pdf"), os.path.join(save_path, "output.pdf"), force_ocr=True, pdf_renderer="sandwich", language="eng")

[verglor]

As mentioned by @alaminkouser setting correct language (ocrmypdf -l eng) fixes the issue so I think this issue can be closed by mentioning this requirement on readme.