Context Tree-Based Image Contour Coding Using a Geometric Prior.

Area of Science:

• Computer Vision
• Image Processing
• Data Compression

Background:

• Efficient encoding of object contours is crucial for advanced image and video compression.
• Current methods face challenges in accurately and efficiently representing contour data.

Purpose of the Study:

• To develop novel algorithms for lossless and lossy compression of detected object contours in images.
• To improve the rate-distortion performance and coding efficiency for contour data.

Main Methods:

• Converting contours into sequences of directional symbols.
• Utilizing a maximum a posterior (MAP) formulation with a variable-length context tree (VCT) for estimating symbol probabilities.
• Designing dynamic programming (DP) algorithms with suffix trees for optimal rate-distortion trade-offs in lossy compression.

Main Results:

• The proposed lossless contour coding algorithm consistently outperforms state-of-the-art context-based schemes.
• The lossy contour coding algorithms demonstrate superior rate-distortion performance compared to existing methods.
• The use of VCT and DP with suffix trees effectively reduces complexity and improves efficiency.

Conclusions:

• The developed methods offer significant improvements in both lossless and lossy contour compression.
• These techniques can facilitate advanced image/video compression by enabling efficient representation of object shapes.
• The study provides a robust framework for optimizing contour encoding in various image processing applications.