GiB: a Game theory Inspired Binarization technique for degraded document images

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Document image binarization is crucial for digitizing historical and degraded documents.
• Image noise, aging, and handling introduce challenges, complicating foreground/background pixel classification.
• Existing binarization methods struggle with significant image degradation.

Purpose of the Study:

• To develop a robust binarization technique for degraded document images.
• To leverage game theory for enhanced local information extraction in pixel classification.
• To improve the accuracy and performance of document image binarization.

Main Methods:

• A novel game theory inspired binarization (GiB) technique was developed.
• A two-player, non-zero-sum, non-cooperative game was designed at the pixel level.
• Preprocessing and post-processing steps were implemented to handle intensity variations and refine results, followed by K-means clustering.

Main Results:

• The GiB method demonstrated superior performance on seven public datasets (DIBCO 2009-14, 2016).
• Experimental results indicate GiB outperforms competing state-of-the-art binarization techniques in most cases.
• The approach effectively handles noise and degradation in document images.

Conclusions:

• The proposed game theory inspired binarization technique offers a significant advancement for processing degraded documents.
• GiB provides a robust and effective solution for accurate document image binarization.
• This method shows promise for applications in digital archiving and document analysis.