Sparse/DCT (S/DCT) two-layered representation of prediction residuals for video coding

Area of Science:

• Video Compression
• Digital Signal Processing
• Information Theory

Background:

• High Efficiency Video Coding (HEVC) is a leading video compression standard.
• Prediction residuals in video coding contain structured information crucial for compression efficiency.
• Existing methods face challenges with side information overhead at higher bit rates.

Purpose of the Study:

• To propose and implement a novel cascaded sparse/DCT (S/DCT) two-layer representation for prediction residuals.
• To improve the Rate-Distortion (R-D) performance of the HEVC standard.
• To address the limitations of sparse representation alone at higher bit rates.

Main Methods:

• An adaptive dictionary is trained to capture featured patterns in residual signals for sparse coding.
• A Discrete Cosine Transform (DCT) representation is cascaded to code the remaining signal.
• The S/DCT approach is integrated into the HEVC framework.

Main Results:

• The S/DCT representation effectively complements sparse coding by handling residual signal energy.
• Experimental results demonstrate superior R-D performance compared to the HEVC reference codec HM5.0.
• The proposed method shows significant improvements under Common Test Conditions.

Conclusions:

• The cascaded S/DCT representation offers enhanced video compression efficiency for HEVC.
• This hybrid approach overcomes the limitations of individual sparse or DCT coding methods.
• The proposed algorithm represents a notable advancement in video coding technology.