Image Quantization Matrix Coding for Lower Code Amount
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image encoding methods, such as HEVC, face challenges in controlling quantization processing of frequency components other than the DC component due to the large size of the CTU, leading to increased code amounts and reduced encoding efficiency.
Innovation Solution
An image encoding apparatus that generates and encodes quantization matrices by associating specific values with DC components and adjacent elements, while omitting associations with other elements, thereby reducing the code amount and allowing finer control over low-frequency quantization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If large orthogonal transformation (16×16 or 32×32 pixels) is used with corresponding quantization matrix, then compression efficiency is improved, but code amount of quantization matrix increases
Solution Approach 1:
The quantization matrix is segmented into multiple regions (first region, second region, third region) with different encoding strategies. The first region uses one value per element for fine control, the second region uses one value per multiple elements for compression, and the third region uses another strategy. This segmentation allows balancing compression efficiency with code amount reduction.
Solution Approach 2:
Different regions of the quantization matrix are assigned different qualities of encoding. The first region (containing important frequency components) receives higher quality encoding with individual value assignment, while the second and third regions use coarser encoding with one value representing multiple elements. This local quality differentiation maintains image quality where needed while reducing overall code amount.
2Measurement precision
If all 256 elements of quantization matrix are encoded, then quantization control precision is improved, but device complexity increases
Solution Approach 1:
The quantization matrix elements are segmented into different groups (first, second, third regions) with different encoding precisions. This segmentation reduces the total number of values that need to be encoded and transmitted, thereby reducing device complexity while maintaining necessary precision in critical regions.
Solution Approach 2:
Instead of encoding all 256 elements with full precision, the invention applies partial action by encoding only certain regions with high precision (one value per element) while using coarser precision for other regions (one value per multiple elements). This partial precision approach reduces complexity while maintaining adequate quantization control.
3Quantity of substance
If up-sampling processing is used to expand 64 elements to 256 elements, then code amount is reduced, but manufacturing precision of quantization matrix decreases
Solution Approach 1:
The quantization matrix is segmented into regions that are directly encoded with specific precision rather than relying solely on up-sampling. The first region is encoded with one value per element, the second region with one value per multiple elements, and the third region with another strategy. This segmentation ensures adequate precision in critical areas while still achieving code amount reduction overall.
Solution Approach 2:
Different regions of the quantization matrix are assigned different encoding qualities. Regions containing important frequency components (first region) receive higher quality encoding with individual value assignment, preventing precision loss that would result from uniform up-sampling of the entire matrix.
Data Source
AI summary
An image decoding apparatus comprises a decoder which decodes data indicating a plurality of values corresponding to a part of a quantization matrix; a generator which derives the plurality of values from the data and generates the matrix; and an inverse quantizing unit which performs the inverse quantizing on an object block using the matrix, wherein, if width or height of the matrix is larger than or equal to a predetermined size, the generating unit generates the matrix by associating a first value among the plurality of values with a first element corresponding to DC component in the matrix, associating a second value with a second element adjacent to the first element, and, for the elements other than the first and second elements, associating each of one or more values with a plurality of elements of the matrix.