Fast Computation of Integer DCT-V, DCT-VIII, and DST-VII for Video Coding

Area of Science:

• Digital signal processing
• Video compression algorithms
• Information theory

Background:

• Joint exploration model (JEM) codecs use various integer transforms (DCT, DST) for improved coding efficiency.
• Fast computation methods for DCT-II and DST-I are established, but less so for DCT-V, DCT-VIII, and DST-VII.

Purpose of the Study:

• To develop fast computation methods for N-point DCT-V and DCT-VIII.
• To optimize these transforms for video coding applications within JEM.

Main Methods:

• Decomposition of DCT-VIII into DST-VII with pre- and post-processing matrices.
• Approximation of integer kernels for DCT-V, DCT-VIII, and DST-VII (N=4, 8, 16, 32) using norm scaling and bit-shift.
• Integration with quantization for video coding.

Main Results:

• Proposed methods effectively reduce matrix operation complexity.
• Little loss in BDBR (Branch-dependent Bit Rate) performance observed.
• Average reduction of 38% in additions and 80.3% in multiplications compared to original JEM.

Conclusions:

• The developed fast computation methods offer significant complexity reduction for JEM video coding.
• These optimizations maintain high coding efficiency, making them suitable for practical implementation.