Chroma Interpolation Control for Low-Bandwidth Image Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods fail to adequately reduce memory bandwidth in video coding and decoding processes, particularly in chroma formats like 4:2:0 and 4:2:2, where the number of color difference samples is smaller than luminance samples, leading to increased memory access and inefficient interpolation processes.
Innovation Solution
An image coding and decoding method that controls interpolation by referring to chroma format information, restricting interpolation for color difference signals in blocks smaller than a predetermined size, and adjusting motion vector values to reduce memory bandwidth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If interpolation filtering process is executed for each block in video coding and decoding, then motion-compensated interpolation accuracy is improved, but memory bandwidth consumption increases
Solution Approach 1:
The patent applies different interpolation filter tap lengths based on block size characteristics. Small blocks use shorter tap length filters (reducing memory access), while large blocks use longer tap length filters (maintaining interpolation accuracy). This local differentiation resolves the contradiction by optimizing memory bandwidth consumption without sacrificing interpolation quality where needed.
Solution Approach 2:
The patent dynamically changes the filter tap length parameter based on block size. By adjusting this parameter, the system adapts the interpolation process to match block characteristics, reducing unnecessary memory accesses for small blocks while maintaining accuracy for large blocks, thus resolving the memory bandwidth versus interpolation accuracy contradiction.
2Measurement precision
If filter with taps longer than two taps is used for color difference interpolation, then color difference interpolation accuracy is improved, but memory bandwidth consumption increases and process restriction becomes difficult
Solution Approach 1:
The patent applies different filter tap lengths to different components (luminance vs. color difference) and different block sizes. For color difference signals in small blocks, shorter tap length filters are used, while luminance signals or large blocks may use longer tap length filters. This component-specific and size-specific differentiation resolves the contradiction by maintaining color difference accuracy only where necessary while reducing memory bandwidth consumption.
Solution Approach 2:
The patent segments the interpolation process by component type (luminance vs. color difference) and block size category. By separating these processing paths, the system can apply optimized filter lengths to each segment, allowing color difference interpolation to use shorter filters in small blocks while maintaining accuracy requirements, thus resolving the memory bandwidth contradiction.
3Measurement precision
If the number of pixels outside the block increases for interpolation, then interpolation quality is improved, but memory bandwidth per pixel increases
Solution Approach 1:
The patent changes the effective filter tap length parameter based on block size. For small blocks, shorter tap length filters are applied, which naturally reduces the number of pixels outside the block that need to be accessed. For large blocks, longer tap length filters can be used since they have fewer edge pixels relative to their total size. This parameter adaptation resolves the contradiction by matching filter complexity to block characteristics.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention aims to reduce a memory bandwidth during image coding. An image coding method includes an acquiring step and a generating step. The acquiring step includes acquiring a reference image. The generating step includes generating a predicted image having resolution larger than that of the reference image, by performing an interpolation to the acquired reference image according to a motion vector for each of the luminance component and the color difference component. The generating step includes generating the predicted image having the color difference component without performing a specific interpolation that means the interpolation processing in which a number of pixels to be accessed in the reference image is large, when a size of a block designated as a unit of the interpolation is smaller than a predetermined threshold value.