Adaptive Sub-sampling of Reference Pixels for Chroma Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding methods, such as HEVC, face challenges in reducing the number of reference pixels while maintaining encoding performance, as the increase in memory capacity does not correspond with improved memory bandwidth, leading to bottlenecks in memory access and higher manufacturing costs and power consumption near calculation cores.
Innovation Solution
The proposed solution involves adaptive-sub-sampling of luma and chroma reference pixels based on the luma intra prediction mode, reducing the number of reference pixels by half while considering image characteristics, and applying this sub-sampling only for the smallest encoding or decoding units to minimize performance degradation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of reference pixels is increased to maintain encoding performance, then encoding performance is preserved, but memory capacity requirements increase and memory bandwidth bottlenecks worsen
Solution Approach 1:
The patent extracts and removes redundant reference pixels from the prediction process. By identifying and eliminating unnecessary reference pixels while preserving essential ones, the system reduces memory capacity requirements without compromising encoding performance. This is achieved through selective reference pixel usage based on prediction mode analysis.
Solution Approach 2:
The patent changes the parameter of reference pixel quantity by dynamically adjusting the number of reference pixels used based on the prediction mode. Different prediction modes utilize different numbers of reference pixels, allowing the system to optimize memory usage while maintaining encoding performance where possible.
2Quantity of substance
If the number of reference pixels is reduced to decrease memory requirements, then memory capacity and power consumption are reduced, but encoding performance degrades
Solution Approach 1:
The patent applies local quality by treating different prediction modes differently in terms of reference pixel requirements. Instead of uniformly reducing reference pixels across all modes, the system selectively reduces them only in modes where it has minimal impact on encoding performance, while maintaining full reference pixel sets in modes where they are critical.
3Quantity of substance
If uniform sub-sampling of reference pixels is applied to reduce their number, then the number of reference pixels is reduced, but encoding performance degrades due to ignoring image characteristics
Solution Approach 1:
The patent implements dynamic reference pixel selection based on the prediction mode. Rather than applying static uniform sub-sampling, the system adaptively determines which reference pixels to use based on the characteristics of each prediction mode, thereby maintaining encoding performance while reducing the number of reference pixels where appropriate.
Data Source
Figure 1
Figure 2
Figure 3(A)~3(B)
AI summary
The number of reference pixels to be referenced in order to reduce redundancy among color components is to be reduced, while suppressing degradation in encoding performance. An intra prediction unit (20A) that performs intra-frame prediction includes an adaptive-sub-sampling luma reference pixel acquisition unit (21A) and an adaptive-sub-sampling chroma reference pixel acquisition unit (22A). In the case where an input image a is an image of a non-YUV420 format and the encoding unit is the smallest CU block, the adaptive-sub-sampling luma reference pixel acquisition unit (21A) sub-samples reference pixels existing neighboring a luma block corresponding to a chroma prediction target block based on a luma intra prediction mode, and acquires a pixel value of each of the sub-sampled reference pixels. In the case where the input image a is an image of a non-YUV420 format and the encoding unit is the smallest CU block, the adaptive-sub-sampling chroma reference pixel acquisition unit (22A) sub-samples reference pixels existing neighboring the chroma prediction target block based on the luma intra prediction mode, and acquires a pixel value of each of the sub-sampled reference pixels.