DNN Deblocking for Block-Wise Image Compression Artifacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in efficiently addressing blocking artifacts in block-wise image compression, particularly due to the inefficiencies in representing less likely intra prediction directions, which result in increased bit usage and reduced compression efficiency.
Innovation Solution
A content-adaptive online training method and apparatus for deblocking in block-wise image compression using a deep neural network (DNN) to determine deblocking parameters, which are updated based on a training process, allowing for effective deblocking of boundary regions in reconstructed blocks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional intra prediction techniques are used with fixed coding structures, then implementation is simple, but compression efficiency is reduced due to increased bit usage for less likely prediction directions
Solution Approach 1:
The patent applies dynamics by making the intra prediction coding structure adaptive rather than fixed. The coding structure changes based on the actual prediction direction used, with more likely directions receiving more bits and less likely directions receiving fewer bits. This dynamic adaptation resolves the contradiction by optimizing compression efficiency for each specific prediction scenario without requiring a completely complex fixed structure for all possible cases.
Solution Approach 2:
The patent changes the parameter allocation (bit usage) based on the prediction direction. By adjusting the number of bits allocated to different intra prediction directions according to their probability of occurrence, the system achieves better compression efficiency. This parameter change approach allows the system to handle less likely directions with fewer bits while maintaining simplicity in the overall coding structure.
2Measurement precision
If more bits are allocated to represent all intra prediction directions equally, then all directions can be represented accurately, but bit usage increases and compression efficiency decreases
Solution Approach 1:
The patent applies local quality by allocating different bit precision to different intra prediction directions based on their likelihood. Instead of using uniform precision for all directions, the system uses higher precision (more bits) for more likely prediction directions and lower precision (fewer bits) for less likely directions. This resolves the contradiction by maintaining adequate representation accuracy for commonly used directions while reducing overall bit usage.
Solution Approach 2:
The patent uses partial action by not allocating full precision to all prediction directions. Instead, it allocates sufficient precision only to the most likely directions that occur frequently in practice, accepting that less likely directions will have reduced representation accuracy. This partial approach significantly reduces bit usage while maintaining overall compression quality.
3Object-affected harmful factors
If a fixed deblocking filter is used, then implementation is straightforward, but blocking artifacts remain in reconstructed blocks
Solution Approach 1:
The patent applies dynamics by making the deblocking filter adaptive rather than fixed. The filter parameters are adjusted based on the local characteristics of the reconstructed blocks, such as the prediction mode and block boundaries. This dynamic adaptation allows the filter to effectively reduce blocking artifacts in different regions while maintaining a relatively simple overall implementation structure.
Solution Approach 2:
The patent applies local quality by applying different deblocking filter strengths and parameters to different boundary regions based on local image characteristics. Instead of using a uniform filter across the entire image, the system adapts the filter parameters locally at each block boundary, which effectively reduces blocking artifacts while keeping the computational complexity manageable through localized processing.
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
Aspects of the disclosure provide a method, an apparatus, and non-transitory computer-readable storage medium for video decoding. The apparatus includes processing circuitry that reconstructs blocks of an image that is to be reconstructed from a coded video bitstream. The processing circuitry decodes first deblocking information in the coded video bitstream including a first deblocking parameter of a deep neural network (DNN) in a video decoder. The first deblocking parameter of the DNN is an updated parameter that has been previously determined by a content adaptive training process. The processing circuitry determines the DNN for a first boundary region comprising a subset of samples in the reconstructed blocks based on the first deblocking parameter included in the first deblocking information. The processing circuitry deblocks the first boundary region comprising the subset of samples in the reconstructed blocks based on the determined DNN corresponding to the first deblocking parameter.