Decoder-Side Inter Prediction Using No-Reference Visual Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding technologies face challenges in efficiently encoding and decoding video sequences due to large data sizes, leading to inefficiencies in storage and transmission, particularly in reducing redundant information without compromising image quality.
Innovation Solution
The implementation of advanced video coding techniques, including intra and inter prediction methods, quadtree+multi-type tree partitioning, and no reference image quality assessment using deep learning, to optimize bitstream compression and decoding processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If advanced prediction methods and partitioning techniques are used, then video encoding efficiency is improved, but device complexity increases
Solution Approach 1:
The video block is divided into multiple sub-blocks using quadtree and multi-type tree partitioning. The current block is first partitioned into four sub-blocks using quadtree, and then each sub-block is further partitioned using multi-type tree into triangular or rectangular regions. This segmentation allows independent prediction for each region, improving encoding efficiency while managing complexity through structured division.
Solution Approach 2:
Different prediction modes are applied to different regions of the video block based on local characteristics. The patent uses asymmetric division where triangular and rectangular regions are created with different orientations, allowing each region to use prediction modes optimized for its specific local structure, thereby improving overall encoding efficiency.
2Loss of substance
If redundant information is reduced through compression, then storage and transmission efficiency is improved, but image quality deteriorates
Solution Approach 1:
The patent changes the prediction parameters by using gradient-based prediction modes (PLANAR, VER, HOR, VD, HD, VU, HU) instead of traditional motion compensation. By calculating gradients in different directions and using them for prediction, the method achieves better compression of redundant information while maintaining image quality through more accurate local prediction.
Solution Approach 2:
The prediction process is made dynamic by adaptively selecting different prediction modes for different regions based on local gradient characteristics. The encoder calculates gradients and dynamically chooses the most appropriate prediction mode for each triangular or rectangular region, allowing the system to adapt to local image characteristics and maintain quality while removing redundancy.
Data Source
AI summary
A decoder decodes, from a bitstream for a current block, an indication of whether decoder-side inter prediction is enabled, and a reconstructed residual block. Based on the decoder-side inter prediction being enabled by the indication and for each respective motion information of a plurality of motion information, the decoder generates a reconstructed block based on: a prediction block generated using the respective motion information, and the reconstructed residual block. For each reconstructed block, the decoder determines a visual quality of the reconstructed block without using the current block as a reference. Based on the visual qualities of the reconstructed blocks, motion information for predicting the current block is determining from the plurality of motion information. The decoder reconstructs the current block based on the determined motion information.