L(infinity) constrained high-fidelity image compression via adaptive context modeling.

Area of Science:

• Digital image processing
• Data compression
• Information theory

Background:

• Existing Differential Pulse Code Modulation (DPCM)-type predictive near-lossless image coders suffer from prediction biases due to quantized prediction residues.
• CALIC (Check All, Keep All) is a state-of-the-art algorithm for image compression, particularly in its near-lossless version.

Purpose of the Study:

• To develop high-fidelity image compression techniques with a strict L(infinity) bound.
• To address and correct prediction biases in existing predictive image coders.
• To improve the performance of the near-lossless CALIC algorithm.

Main Methods:

• Proposed practical adaptive context modeling techniques.
• Corrected prediction biases caused by quantizing prediction residues.
• Integrated these techniques into the near-lossless CALIC algorithm.

Main Results:

• Achieved a peak signal-to-noise ratio (PSNR) increase of 1 dB or more.
• Reduced bit rates by 10% or more.
• Obtained competitive PSNR results against L(2)-based wavelet coders at bit rates around 1.25 bits per pixel (bpp) and higher, with a significantly smaller L(infinity) bound.

Conclusions:

• The proposed adaptive context modeling effectively corrects prediction biases in near-lossless image compression.
• The enhanced CALIC algorithm offers superior performance in terms of PSNR and bit rate efficiency.
• The method provides a competitive alternative to L(2)-based wavelet coders, especially when a tight L(infinity) bound is required.