off-nutrition-table-extractor icon indicating copy to clipboard operation
off-nutrition-table-extractor copied to clipboard

Published 20 hours ago • verified publisher icon openfoodfacts

Develop a better image preprocessing algorithm.

Open sgrpanchal31 opened this issue 6 years ago • 0 comments

Currently, we are using the following filters before sending the images for OCR: RGB -> Grayscale -> GaussianBlur -> Grayscale -> RGB The problem we are facing is that some of the bold text is not been able to detect by OCR. Also, some of the images with non-black backgrounds are undetectable. You can find the algorithm in the file process.py file under the function name preprocess_for_ocr.

sgrpanchal31 avatar Aug 19 '18 17:08 sgrpanchal31