Image Decoding Context Models for Inter Prediction Bit Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image encoders/decoders face inefficiencies in bit allocation and entropy coding due to mismatched sequences of inter prediction modes, leading to increased bit overhead and performance loss.
Innovation Solution
The method involves arranging inter prediction mode information in order of occurrence frequency and combining similar modes to reduce bits, using context models based on neighboring block prediction information for entropy encoding/decoding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If inter prediction modes are expressed using a pre-determined sequence, then the encoding structure is simple, but binary bits are inefficiently allocated causing performance loss
Solution Approach 1:
The patent applies dynamics by changing the inter prediction mode sequence from a fixed pre-determined order to a dynamic order based on occurrence frequency. The encoder determines the sequence adaptively according to how frequently each mode appears in the video content, allowing the most frequent modes to be encoded with fewer bits while maintaining a manageable encoding structure.
2Adaptability or versatility
If inter prediction modes are divided into different indicators, then various inter prediction information can be expressed, but bit overhead increases and probability-based entropy coding performance deteriorates
Solution Approach 1:
The patent merges multiple inter prediction modes into unified indicators based on their functional similarities. Instead of using separate indicators for each mode, the patent combines modes with similar characteristics into single indicator categories, reducing the total number of indicators and thereby decreasing bit overhead while maintaining the ability to express diverse prediction information.
Data Source
AI summary
Disclosed herein is an image decoding method. The method of decoding an image comprises obtaining prediction information and residual signal information for a current block from a bitstream and reconstructing, based on the obtained prediction information and the residual signal information, the current block, wherein the prediction information is obtained through entropy decoding, and wherein a context model used for the entropy decoding of the prediction information is determined based on a prediction mode of a neighboring block of the current block.