Adaptive In-Loop Filtering for Lower-Buffer Image Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image encoding and decoding technologies face challenges in managing the size of buffers required for storing residual and prediction samples during in-loop filtering, particularly with advancements in hardware and AI technologies.
Innovation Solution
An image decoding and encoding method that determines a filter index for in-loop filtering based on representative values of current and neighboring blocks, allowing adaptive selection of in-loop filter sets to optimize buffer usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If in-loop filtering is performed using residual samples or prediction samples in addition to reconstructed samples, then image quality is enhanced, but buffer size for storing samples increases
Solution Approach 1:
The patent divides the filtering process into multiple stages: reconstruction filtering using only reconstructed samples, and detail enhancement filtering using residual or prediction samples. This segmentation allows the system to achieve high image quality while managing buffer requirements by processing different sample types in separate stages rather than requiring all samples simultaneously.
Solution Approach 2:
The patent performs reconstruction filtering as a preliminary action before detail enhancement filtering. By first applying in-loop filtering to reconstructed samples and then optionally enhancing with residual/prediction samples, the system prepares the image data in advance, reducing the need to hold all sample types in buffer simultaneously.
2Adaptability or versatility
If multiple in-loop filter sets are maintained for different block types, then filtering adaptability is improved, but device complexity increases
Solution Approach 1:
The patent introduces dynamic filter selection based on block characteristics. Instead of maintaining multiple static filter sets, the system dynamically selects and applies appropriate filtering parameters based on the current block type (intra-prediction, inter-prediction, merge mode, etc.), achieving adaptability without proportionally increasing device complexity.
Solution Approach 2:
The patent changes filtering parameters adaptively based on block type rather than maintaining separate filter sets. By modifying filter strength, kernel size, or application conditions based on prediction mode and block characteristics, the system achieves versatile filtering behavior while using a unified filter implementation.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided are an image decoding method and apparatus and an image encoding method and apparatus which involve obtaining a first representative value of a current block with respect to residual samples or prediction samples of the current block, and a second representative value of each of neighboring blocks with respect to residual samples or prediction samples of each of the neighboring blocks of the current block, obtaining a filter index indicating one in-loop filter set among a plurality of in-loop filter sets, based on the first representative value and the second representative value, and performing in-loop filtering on the current block, based on the one in-loop filter set determined by the filter index.