Context-based multiscale classification of document images using wavelet coefficient distributions.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Accurate segmentation of document images is crucial for information retrieval and analysis.
• Existing methods often struggle with class boundaries and overall efficiency.

Purpose of the Study:

• To develop an adaptive algorithm for segmenting document images into four distinct classes: background, photograph, text, and graph.
• To improve classification accuracy and processing speed through multiscale analysis and context accumulation.

Main Methods:

• Feature extraction based on distribution patterns of wavelet coefficients in high-frequency bands.
• Adaptive multiscale classification, processing images at various resolutions.
• Incorporation of accumulated context information to refine classification.

Main Results:

• The algorithm successfully segments document images into the four specified classes.
• The multiscale nature enables accurate classification at boundaries and efficient overall processing.
• Accumulated context information demonstrably improves classification accuracy.

Conclusions:

• The developed algorithm provides an effective and efficient solution for document image segmentation.
• Its adaptive multiscale and context-aware approach offers significant advantages over traditional methods.