harraj2015ocr.md

OCR accuracy improvement on document images through a novel pre-processing approach

Abdeslam El Harraj, Naoufal Raissouni

Browse

@article{article,
author = {Harraj, Abdeslam and Raissouni, Naoufal},
year = {2015},
month = {09},
pages = {},
title = {OCR Accuracy Improvement on Document Images Through a Novel Pre-Processing Approach},
volume = {6},
journal = {Signal & Image Processing : An International Journal},
doi = {10.5121/sipij.2015.6401}
}

Pipeline

Receipt detection	Receipt localization	Receipt normalization	Text line segmentation	Optical character recognition	Semantic analysis
❌	❌	✔️	❌	❌	❌

Receipt normalization

• Illumination adjustment, grayscale conversion, Un-sharp masking and optimized adaptive thresholding for binarization

  

• Illumination adjustment - CLAHE as a powerful contrast enhancement

  • Brightness Equalization - CLAHE is based on Adaptive Histogram Equalization (AHE). CLAHE can limit the noise whereas enhancing the contrast
  • Brightness estimation - Y’UV is more suitable for our case to estimate the brightness

• Grayscale conversion

• Un-sharp masking

• Otsu thresholding - Cleaning and whitening document background

Notes

• nonparametric and unsupervised method to compensate for undesirable document image distortions aiming to optimally improve OCR accuracy