Robust Combined Binarization Method of Non-Uniformly Illuminated Document Images for Alphanumerical Character...

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Image binarization is crucial for image analysis, but simple methods fail with poor illumination or degraded quality.
• Existing binarization techniques lack universality, necessitating specialized approaches for diverse image types like historical documents or mobile captures.
• Optical Character Recognition (OCR) accuracy is heavily dependent on effective pre-processing, including binarization, particularly for challenging image conditions.

Purpose of the Study:

• To develop an improved image binarization method that enhances Optical Character Recognition (OCR) accuracy.
• To address the limitations of existing binarization techniques for non-uniformly illuminated and degraded document images.
• To propose a robust combined approach integrating advantages of various binarization strategies.

Main Methods:

• A novel robust combined measure for image binarization was developed.
• The approach integrates recent techniques, including entropy filtering and multi-layered region analysis.
• Experiments were conducted on the WEZUT OCR Dataset, comprising 176 non-uniformly illuminated document images.

Main Results:

• The proposed robust combined binarization method significantly increased OCR accuracy.
• The approach demonstrated effectiveness on challenging, non-uniformly illuminated document images.
• Experimental validation on the WEZUT OCR Dataset confirmed the method's validity and usefulness.

Conclusions:

• Robust combined measures offer a superior solution for image binarization compared to single methods.
• The developed binarization technique is particularly beneficial for improving OCR performance on difficult document images.
• This work contributes to advancing OCR technology for real-world applications, including mobile document capture.