Entropy-based evaluation of context models for wavelet-transformed images

Area of Science:

• Information Theory
• Digital Image Processing
• Computer Vision

Background:

• Entropy quantifies message uncertainty and sets theoretical limits for data compression.
• Wavelet-based image coding utilizes context models to improve compression efficiency.
• Evaluating the performance of these context models is crucial for codec optimization.

Purpose of the Study:

• To define an entropy-based measure for evaluating context models in wavelet-based image coding.
• To establish the maximum achievable performance for each context model.
• To predict the coding rates of practical image coding systems with high precision.

Main Methods:

• Developed an entropy-based metric tailored to modern coding system mechanisms.
• Applied the metric to evaluate four established context models.
• Conducted experiments using diverse image types, coding rates, and transform strategies.

Main Results:

• The proposed measure accurately predicts the performance of context models.
• Identified specific coding conditions where widely-used context models are suboptimal.
• Demonstrated the variability in context model adequacy across different scenarios.

Conclusions:

• The developed entropy-based measure effectively assesses context model performance in wavelet image coding.
• Certain conventional context models may not be universally optimal.
• Findings can guide the design of simpler and more efficient wavelet image codecs.