Transforms for Intra Prediction Residuals Based on Prediction Inaccuracy Modeling.

Area of Science:

• Computer Vision
• Digital Signal Processing
• Video Compression

Background:

• Directional intra prediction is crucial for reducing spatial redundancy in video and image coding.
• Intra prediction residuals are typically encoded using transforms.
• Accurate prediction direction is challenging to achieve in practical video coding systems.

Purpose of the Study:

• To develop novel transforms specifically for directional intra prediction residuals.
• To address the inaccuracies in prediction direction inherent in practical systems.
• To enhance the energy compaction property of transforms for these residuals.

Main Methods:

• Estimating residual covariance based on prediction direction inaccuracies.
• Proposing a class of transforms derived from the estimated covariance function.
• Evaluating the energy compaction performance of the proposed transforms.

Main Results:

• The proposed transforms demonstrate superior energy compaction for directional intra prediction residuals.
• A significantly smaller number of transform coefficients are required to preserve the same energy.
• The method effectively handles residuals arising from imperfect prediction directions.

Conclusions:

• The developed transforms offer an efficient solution for encoding directional intra prediction residuals.
• This approach leads to improved compression efficiency in video coding systems.
• The method provides better energy preservation with fewer coefficients, optimizing data transmission.