Inverse LFNST Coefficient Handling Under Worst-Case Decode Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies, such as VVC, hardwire the zeroing out of transform coefficients to meet worst-case multiplication criteria, leading to unnecessary reduction in coding quality.
Innovation Solution
Determine the number of allowed non-zero transform coefficients as a function of block size and transform dimensions to selectively apply inverse low-frequency non-separable transforms (LFNST) and separable transforms, optimizing coefficient reconstruction while adhering to worst-case multiplication constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If transform coefficients are zeroed out to meet worst-case multiplication criteria, then device complexity is reduced, but coding quality deteriorates
Solution Approach 1:
The patent makes the number of non-zero transform coefficients dynamic rather than static. The encoder determines the actual number of non-zero coefficients based on the specific block content and signaling requirements, allowing the system to adapt between different complexity levels and quality requirements rather than being constrained by a fixed worst-case assumption.
Solution Approach 2:
The patent changes the parameter of transform coefficient handling from a fixed zeroing approach to a flexible signaling approach. By introducing syntax elements that indicate the actual number of non-zero coefficients, the system can adjust the degree of coefficient retention based on coding conditions, thereby improving quality without excessively increasing complexity.
2Device complexity
If the number of non-zero transform coefficients is limited, then worst-case multiplication count is reduced, but coding efficiency deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the encoder signals the actual number of non-zero transform coefficients to the decoder. This feedback allows the decoder to know exactly how many coefficients to process, eliminating unnecessary multiplications for zero coefficients while maintaining all non-zero coefficients for optimal reconstruction quality.
Solution Approach 2:
Instead of applying a fixed partial action (zeroing out a predetermined number of coefficients), the patent uses partial action adaptively - only zeroing out coefficients that are actually zero based on the encoded data. This avoids both excessive zeroing (which would harm quality) and insufficient zeroing (which would increase complexity unnecessarily).
Data Source
AI summary
A video decoder can be configured to determine a number of allowed non-zero coefficients for a block of video data based on a size of the block; obtain a set of dequantized coefficients for the block, wherein the set of dequantized coefficients comprises a first subset of dequantized coefficients that includes non-zero dequantized coefficients and a second subset of dequantized coefficients that includes all zero coefficients, wherein a number of coefficients in the first subset of dequantized coefficients is equal to the number of allowed non-zero coefficients for the block of video data; apply an inverse low-frequency non-separable transform (LFNST) to the first subset of dequantized coefficients to determine a first intermediate subset of coefficients; and apply an inverse separable transform to the first intermediate subset of coefficients and at least a portion of the second subset of coefficients to determine a block of reconstructed residual values.