Low latency secondary transforms for intra/inter prediction residual

Area of Science:

• Digital video compression
• Image and video processing
• Signal processing

Background:

• Block-based video codecs commonly use Discrete Cosine Transform (DCT) for residue coding.
• DCT's effectiveness can be further enhanced for specific prediction scenarios.

Purpose of the Study:

• To propose and evaluate a novel transform scheme that improves video coding efficiency.
• To introduce a secondary transform applied after DCT for enhanced residue coding.

Main Methods:

• A secondary transform is applied to lower-frequency DCT coefficients (K=4 or 8) for blocks of size 8x8 and larger.
• The scheme is compatible with separable horizontal and vertical transform video codecs.
• The transform is applied selectively based on prediction direction (intra) or transform unit location (inter).

Main Results:

• Significant BD-rate improvements achieved compared to conventional DCT-based coding.
• Low computational complexity due to application on a subset of coefficients, especially for large blocks (e.g., 32x32).
• No additional signaling or Rate-Distortion (R-D) search required.

Conclusions:

• The proposed secondary transform scheme offers substantial coding gains in video compression.
• The technique is hardware-friendly, enabling low-latency implementation.
• It provides an efficient enhancement to existing DCT-based video coding standards.