Lossless Image Compression Using Context-Dependent Linear Prediction Based on Mean Absolute Error Minimization

Area of Science:

• Computer Vision
• Image Processing
• Data Compression

Background:

• Lossless image compression is crucial for preserving image fidelity.
• Existing methods like JPEG-LS offer compression but can be improved for efficiency and speed.
• Adaptive modeling based on image characteristics is key to enhancing compression ratios.

Purpose of the Study:

• To develop a lossless image compression method with fast decoding.
• To enhance compression efficiency by adapting encoder parameters to image content.
• To evaluate the proposed method against established compression standards.

Main Methods:

• Data modeling using linear and nonlinear prediction, enhanced by a context-dependent constant component removal block.
• Prediction error encoding via two-stage compression: adaptive Golomb and binary arithmetic coding.
• Implementation of an author's context-switching algorithm for tailored prediction models across image areas.

Main Results:

• Achieved high compression efficiency by adapting prediction models to specific image regions.
• Demonstrated a 9.1% improvement in average file bit rate compared to JPEG-LS across a test base.
• Analyzed the trade-offs between individual encoder parameters, compression efficiency, and encoding time.

Conclusions:

• The proposed lossless image compression method offers superior efficiency and fast decoding.
• Context-switching algorithms significantly improve compression by tailoring models to image data.
• The method presents a competitive alternative to existing lossless image compression techniques.