Quantization Matrix Scaling for HEVC Storage Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increased data size requirements for quantization matrices in High Efficiency Video Coding (HEVC) systems pose challenges for storage and transmission, particularly when user-defined matrices are used, as they significantly impact hardware implementation and bitrate.
Innovation Solution
Deriving larger quantization matrices from initial 4×4 and 8×8 matrices through spatial-frequency up-sampling and linear interpolation, reducing the need to store and transmit larger matrices explicitly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If larger quantization matrices (16×16, 32×32) are explicitly stored and transmitted in HEVC systems, then the video coding standard supports larger transform sizes, but the data size requirement for quantization matrices increases substantially (more than 18 times compared to MPEG2/H264/AVC)
Solution Approach 1:
The patent creates scaled copies of the base 8×8 quantization matrix to generate 16×16 and 32×32 quantization matrices. Instead of storing separate full-sized matrices, the system stores one base matrix and generates larger versions by repeating and scaling its values, dramatically reducing storage requirements while maintaining the ability to support multiple transform sizes.
Solution Approach 2:
The base 8×8 quantization matrix serves multiple functions by being scaled to generate quantization matrices of different sizes (16×16, 32×32). A single stored matrix enables support for various transform block sizes through scaling operations, making the quantization matrix system multi-functional and reducing the total number of matrices that need to be stored and transmitted.
2Reliability
If user-defined quantization matrices are used in HEVC, then the system can optimize for specific video content, but large quantization matrices consume sizeable precious bitrate
Solution Approach 1:
The patent transmits only the base 8×8 user-defined quantization matrix and generates larger matrices (16×16, 32×32) through scaling operations at the decoder. This copying approach reduces the bitrate consumption for transmitting user-defined matrices from what would be required to transmit all possible sizes explicitly, while still allowing optimization for specific video content through the user-defined base matrix.
3Adaptability or versatility
If default quantization matrices for all transform sizes are stored, then the system has complete coverage for all block sizes, but the memory requirement increases from 160 bytes (MPEG2/H264/AVC) to 2048 bytes (HEVC)
Solution Approach 1:
The patent stores only the base 8×8 default quantization matrix and generates 16×16 and 32×32 matrices by scaling. This copying strategy reduces memory requirements from storing all matrices explicitly (2048 bytes) to storing just the base matrix and generating others through scaling operations, while maintaining complete block size coverage.
Solution Approach 2:
The patent combines the functionality of multiple quantization matrices (8×8, 16×16, 32×32) into a single base 8×8 matrix that can generate all required sizes through scaling. This merging approach consolidates what would be separate storage requirements into one compact representation, reducing overall memory usage while maintaining adaptability to different block sizes.
Data Source
AI summary
A method and system process transform blocks according to quantization matrices in a video coding system. The method includes obtaining an initial quantization matrix having a first width and a first height, obtaining a derived quantization matrix having a second width and a second height, wherein the second width is different from the second height, and the derived quantization matrix is derived from the initial quantization matrix, receiving a transform block having a block size, where the transform block is associated with a picture, and selecting the initial quantization matrix or the derived quantization matrix for processing transform coefficients of the transform block according to the block size.