DNN In-Loop Filtering for Image Quality and Complexity Trade-offs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image compression technologies face challenges in maintaining image quality due to quantization errors, and existing in-loop filtering methods may not adequately address these issues, particularly in encoding and decoding processes.
Innovation Solution
The implementation of a deep neural network (DNN) filter model for in-loop filtering, which determines the most suitable filtering operations based on content type and compression strength, applying deblocking filtering, sample adaptive offset, and adaptive loop filtering to reconstructed data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional in-loop filtering methods are used, then encoding complexity is reduced, but image quality and quantization error compensation are insufficient
Solution Approach 1:
The patent segments the filtering process into multiple specialized filters (deblocking filter, sample adaptive offset filter, adaptive loop filter) that operate sequentially on different aspects of quantization errors. Each filter targets specific types of artifacts, allowing complex image quality improvement through multiple simple, manageable filtering stages rather than a single complex filter
Solution Approach 2:
The patent applies in-loop filtering early in the decoding process, before final image output and display. By performing deblocking, sample adaptive offset, and adaptive loop filtering on reconstructed data before it is used for subsequent processing or displayed, the system proactively compensates for quantization errors before they propagate through the system or become visible to users
2Manufacturing precision
If strong filtering is applied to compensate quantization errors, then image quality improves, but processing time and computational load increase
Solution Approach 1:
The patent applies different filtering strengths and types to different regions of the image based on local characteristics. The adaptive loop filter and sample adaptive offset adjust their filtering intensity according to local variance, edge detection, and content complexity, applying strong filtering only where quantization errors are most apparent while using minimal filtering in smooth regions, thereby optimizing the balance between quality improvement and processing time
3Manufacturing precision
If multiple filtering operations are applied, then quantization error compensation improves, but device complexity and computational resources increase
Solution Approach 1:
The patent combines multiple filtering operations (deblocking, sample adaptive offset, adaptive loop filtering) into a unified in-loop filtering framework that processes reconstructed data through a coordinated sequence of filters. These filters share common computational resources, data structures, and control mechanisms, allowing multiple filtering operations to be performed with less incremental complexity than if each filter operated independently
Data Source
AI summary
Provided is in-loop filtering technology using a trained deep neural network (DNN) filter model. An image decoding method according to an embodiment includes receiving a bitstream of an encoded image, generating reconstructed data by reconstructing the encoded image, obtaining information about a content type of the encoded image from the bitstream, determining a deep neural network (DNN) filter model trained to perform in-loop filtering by using at least one computer, based on the information about the content type, and performing the in-loop filtering by applying the reconstructed data to the determined DNN filter model.