Neural Network Filtering of Image Blocks for Coding Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multimedia data processing methods suffer from low accuracy in image filtering, resulting in poor filtering effects and reduced coding quality due to the distortion introduced by quantization during the coding process.
Innovation Solution
A multimedia data processing method that utilizes a neural network-based image filtering processor to enhance filtering accuracy by incorporating target coding and decoding information, leveraging associated image blocks to improve the filtering process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional filtering modes are used on reconstructed images, then the processing complexity is low, but the filtering accuracy is poor resulting in distorted images
Solution Approach 1:
The patent introduces an associated image block as an intermediary element that bridges the gap between the target image block and the filtering process. This associated image block contains high-quality reference information that guides the neural network-based filtering, enabling accurate reconstruction without requiring complex manual filtering rules. The intermediary block serves as a mediator that transfers quality information from reference images to the target block being filtered.
Solution Approach 2:
The patent replaces traditional mechanical filtering rules (hand-crafted filtering algorithms) with a neural network-based image filtering processor. This substitution allows the system to learn optimal filtering strategies from data rather than relying on predefined rules, significantly improving filtering accuracy while the neural network handles the complexity internally.
2Productivity
If quantization operations are performed during coding, then the coding efficiency is improved, but the reconstructed image quality deteriorates due to information loss
Solution Approach 1:
The patent performs preliminary actions by determining and utilizing an associated image block before the filtering process. This associated image block is prepared in advance and contains high-quality reference information that compensates for the information lost during quantization. By having this reference information ready beforehand, the system can effectively guide the neural network to recover quantization artifacts without requiring re-coding.
Solution Approach 2:
The patent converts the harmful effect of quantization (information loss and distortion) into a benefit by using the quantized coding and decoding information as input features for the neural network. The neural network learns to recognize and correct quantization artifacts, transforming the previously harmful quantization process into an opportunity for intelligent error correction and quality enhancement.
3Measurement precision
If more information is used for filtering the target image block, then the filtering accuracy increases, but the information processing load increases
Solution Approach 1:
The patent segments the information processing by dividing it into distinct components: the target image block, the associated image block, and the coding and decoding information. Each segment is processed separately and then integrated by the neural network. This segmentation allows the system to efficiently manage information flow and computational resources, processing only the necessary segments rather than handling all possible image data simultaneously.
Data Source
Figure 1~3
Figure 4~6
Figure 7~8b
AI summary
Disclosed in the embodiments of the present application are a multimedia data processing method and apparatus, and a device, a storage medium and a program product. The method comprises: determining a first associated image block associated with a target image block to be filtered in multimedia data; acquiring the first associated image block and target encoding and decoding information associated with the target image block; and according to the first associated image block and the target encoding and decoding information, performing filtering processing on the target image block by means of an image processing filter based on a neural network to obtain a filtered image block corresponding to the target image block.