Reduced Transform Coefficient Matrix for Video Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of large forward transforms in modern video coding standards, such as HEVC, increases computational resources and processing time, limiting coding efficiency and speed.
Innovation Solution
Conditionally zeroing out transform matrix coefficients to create a reduced matrix with a non-zero inner region and zero outer region, reducing the computational resources required for video encoding while minimizing content loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If large transform matrix sizes (16×16 and 32×32) are used in HEVC, then coding efficiency is improved, but computational complexity and processing time increase
Solution Approach 1:
The transform matrix is divided into multiple sub-matrices or blocks. Instead of performing a single large 16×16 or 32×32 transform, the residual block is partitioned into smaller regions, and transforms are applied to each region separately. This segmentation reduces the computational complexity of each individual transform operation while maintaining overall coding efficiency through selective application.
Solution Approach 2:
Different transform sizes and types are applied to different regions of the residual block based on local characteristics. High-frequency regions may use smaller transforms or different transform types compared to low-frequency regions. This local adaptation optimizes the balance between coding efficiency and computational complexity by applying appropriate transform complexity only where needed.
2Productivity
If large transform matrix sizes (16×16 and 32×32) are used in HEVC, then coding efficiency is improved, but processing speed decreases
Solution Approach 1:
The large transform operation is segmented into multiple smaller transform operations that can be executed more quickly. By dividing the 16×16 or 32×32 transform into smaller blocks, the processing time for each block is reduced, and the overall processing speed increases while maintaining the benefits of larger transform sizes through selective application.
Solution Approach 2:
Instead of applying large transforms to all residual blocks, the patent applies them selectively only to blocks where they provide significant coding efficiency benefits. For blocks where smaller transforms are sufficient, partial action is taken, reducing the overall processing load and increasing average processing speed while maintaining coding efficiency where it matters most.
3Measurement precision
If full transform matrix coefficients are used, then transform accuracy is maintained, but computational resources increase
Solution Approach 1:
The patent applies different levels of transform precision to different regions of the residual block. High-frequency regions that require higher precision for accurate reconstruction use full transform coefficients, while low-frequency regions use reduced precision or fewer coefficients. This local quality approach maintains transform accuracy where it is most critical while reducing computational resources in regions where full precision is less important.
Solution Approach 2:
The transform parameters such as matrix size, coefficient precision, and transform type are dynamically changed based on the characteristics of the residual block. This allows the system to adapt the transform accuracy and computational resource usage to match the actual content being encoded, maintaining high accuracy when needed while reducing resource consumption for simpler content.
Data Source
AI summary
Systems and methods for low complexity forward transforms using zeroed-out coefficients are described herein. One aspect of the subject matter described in the disclosure provides a video encoder comprising a memory configured to store a video block. The video encoder further comprises a processor in communication with the memory. The processor is configured to determine a full power value of the video block. The processor is further configured to determine a reduced transform coefficient matrix, wherein the reduced transform coefficient matrix comprises an inner region of zero or non-zero values of the same inner region of a full transform coefficient matrix and an outer region of zero values, wherein the reduced transform coefficient matrix and the full transform coefficient matrix have the same size. The processor is further configured to determine a partial power value of the video block using the reduced transform coefficient matrix. The processor is further configured to transform the video block from a pixel domain to a coefficient domain using the reduced transform coefficient matrix based on the full power value and partial power value. The processor is further configured to encode the transformed video block.