Implicitly Selected Transform for AVS3

Area of Science:

• Digital video compression
• Information theory
• Signal processing

Background:

• Transform coding is vital for video compression by reducing data redundancy.
• Multiple transform cores improve coding efficiency but incur overhead for signaling transform types.
• Implicitly Selected Transform (IST) offers a potential solution to reduce this overhead.

Purpose of the Study:

• To investigate the statistical properties of transform coefficients for improved video coding.
• To propose an efficient IST method that minimizes signaling overhead.
• To enhance the performance of the AVS3 video coding standard.

Main Methods:

• Statistical analysis of transform coefficients.
• Development of a parity adjustment scheme for IST.
• Integration with rate-distortion optimization (RDO) quantization.
• Proposal of size-based and number-based IST combination methods.
• Application of a restriction region to reduce computational complexity.

Main Results:

• Theoretical analysis confirms the superiority of implicit signaling over explicit methods for IST.
• The proposed parity adjustment scheme with RDO achieves significant BD-Rate savings (1.76% AI, 0.76% RA on average).
• The method demonstrates better coding gain compared to explicit signaling methods.
• Negligible impact on decoding time was observed.
• The proposed method was adopted into the AVS3 video coding standard.

Conclusions:

• The developed parity adjustment scheme for IST is effective in reducing video coding redundancy and overhead.
• The optimized IST approach provides superior coding performance and efficiency.
• The method's successful integration into AVS3 validates its practical applicability and effectiveness.