Adaptive Transform Matrix Selection in Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding methods face challenges in efficiently compressing higher resolution videos due to increased bandwidth demands and computational complexity, particularly in handling different prediction modes and block sizes effectively.
Innovation Solution
The implementation of multiple transforms, including Discrete Cosine Transforms (DCT) and Discrete Sine Transforms (DST), along with transform skip modes and training process-based transforms, is used to optimize video coding by selecting appropriate transform matrices based on block size, prediction modes, and color components, enabling efficient coding and decoding operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple transform matrices are used to improve compression performance, then video coding efficiency is improved, but device complexity increases
Solution Approach 1:
The patent implements dynamic transform selection where the encoder and decoder adaptively choose from multiple transform matrix sets based on block characteristics. The system dynamically switches between different transform types (e.g., DCT, DST, DRT) and matrix sets (first set for intra prediction, second set for inter prediction) depending on the prediction mode and block size, optimizing compression efficiency while managing complexity through conditional selection rather than exhaustive processing
Solution Approach 2:
The patent applies different transform matrices to different regions and block types within the video stream. Specifically, it uses a first set of transform matrices for intra-predicted blocks and a second set for inter-predicted blocks, and further differentiates between block sizes (e.g., 4×4, 8×8, 16×16). This localized application of appropriate transforms improves compression performance for each block type while avoiding the complexity of applying all transforms to all blocks
2Productivity
If transform operations are applied to all video blocks, then compression performance is improved, but computational complexity increases
Solution Approach 1:
The patent changes the parameters of the transform operation by selecting from multiple matrix sets based on block characteristics. It introduces a low-frequency non-separable transform index that determines which transform matrix set to use, and applies zero-out operations that set certain coefficients to zero based on frequency content. These parameter changes enable efficient compression by adapting the transform to the specific characteristics of each block, improving performance while controlling computational load through selective application
3Productivity
If zero-out operations are applied to reduce coefficients, then coding efficiency is improved, but manufacturing precision requirements increase
Solution Approach 1:
The patent applies zero-out operations selectively to specific coefficient ranges rather than all coefficients. It uses a zero-out range parameter that determines which coefficients are set to zero based on the transform type and block characteristics. This partial application of zero-out operations achieves coding efficiency improvements by eliminating redundant high-frequency coefficients while maintaining precision requirements only where necessary for visual quality
Data Source
AI summary
Devices, systems and methods for digital video coding, which includes methods for transform design, are described. In a representative aspect, a method of video processing includes performing a conversion between a current video block of a video and a bitstream of the video based on a rule, wherein the rule specifies that a selection of a transform matrix set for performing a transform operation during the conversion is based on a low-frequency non-separable transform index indicated in the bitstream, wherein the rule specifies that the transform operation includes, during an encoding operation, coding the current video block into the bitstream by applying a forward transform on residual values of the current video block, or wherein the rule specifies that the transform operation includes, during a decoding operation, generating from the bitstream the current video block by applying an inverse transform to scaled coefficients indicated in the bitstream.