Inter-Prediction Offset Refinement for Video Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression techniques, such as H.264/AVC and HEVC, face challenges in achieving high encoding and decoding efficiency due to limitations in motion estimation accuracy, particularly in the conventional merge mode.
Innovation Solution
The proposed solution involves an inter-prediction method that extracts merge candidate flags and offset information from a bitstream, selects a merge candidate from a list of neighboring blocks, applies offset information to the motion vector of the selected merge candidate to determine the motion vector of the current block, and generates a prediction block based on this motion vector.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional merge mode is used for motion estimation, then device complexity is reduced, but motion estimation accuracy deteriorates
Solution Approach 1:
The patent segments the motion estimation process into two parts: first selecting a merge candidate from neighboring blocks using simple merge mode, then applying offset information to refine the motion vector. This segmentation allows the system to maintain low complexity while improving accuracy through the additional offset adjustment step.
Solution Approach 2:
The patent performs preliminary action by first selecting a merge candidate from the merge candidate list before applying offset information. This preliminary selection provides a baseline motion vector that is then refined, allowing the system to start with a simple approximation and improve it systematically without大幅增加 complexity.
2Measurement precision
If offset information is added to merge mode to improve accuracy, then motion estimation accuracy is improved, but the number of bits required to signal offset information increases
Solution Approach 1:
The patent applies partial action by using limited offset information rather than full motion vector refinement. The offset values are constrained to specific ranges and are only applied when beneficial, avoiding the transmission of excessive information while still achieving improved accuracy over conventional merge mode.
Solution Approach 2:
The patent changes the parameter representation by using offset values relative to the selected merge candidate's motion vector. This parameter transformation allows the system to encode only the difference (offset) rather than the complete motion vector, reducing the number of bits required while maintaining accuracy.
3Manufacturing precision
If video resolution and frame rate are increased, then image quality is improved, but the amount of data to be encoded increases
Solution Approach 1:
The patent uses feedback by selecting merge candidates from neighboring blocks that have similar motion characteristics. This local feedback mechanism allows the encoder to exploit temporal and spatial redundancy in high-resolution, high-frame-rate video, reducing the amount of new information that needs to be encoded while maintaining image quality.
Solution Approach 2:
The patent employs copying by reusing motion vectors from neighboring blocks (merge candidates) instead of encoding completely new motion information for each block. This copying approach significantly reduces the data amount required for high-resolution video while maintaining accurate motion representation.
Data Source
AI summary
Disclosed are an inter-prediction method and an image decoding apparatus using the same. According to an embodiment of the present invention, there is provided an inter-prediction method including extracting a merge candidate flag and offset information from a bitstream, selecting a merge candidate corresponding to the extracted merge candidate flag from a merge candidate list including neighboring blocks of a current block as merge candidates, deciding a motion vector of the current block by applying the offset information to a motion vector of the selected merge candidate, and generating, as a prediction block of the current block, a block indicated by the motion vector of the current block in a reference picture referenced by the selected merge candidate.