Optical Flow Motion Vector Bi-directional Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video compression technologies face challenges in efficiently processing next-generation video content with high spatial resolution, high frame rate, and high dimensionality, requiring improved coding tools to enhance memory storage, memory access rate, and processing power.
Innovation Solution
The method involves using an optical flow motion vector for bi-directional prediction by applying weights to the gradient components of an optical flow predictor in generalized bi-prediction, incorporating adaptive weight factors to improve predictor efficiency without additional motion vector transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If optical flow motion vector is used for bi-directional prediction, then compression performance is improved, but computational complexity increases
Solution Approach 1:
The patent divides the current block into multiple sub-blocks and performs separate optical flow-based bi-directional prediction for each sub-block. This segmentation allows the computationally intensive optical flow calculation to be distributed across smaller regions, improving compression performance through more precise motion modeling while managing computational complexity through localized processing.
Solution Approach 2:
The patent performs preliminary optical flow estimation and gradient calculation on reference pictures before the actual prediction process. By pre-computing motion vectors and gradient maps from reference frames, the method reduces real-time computational burden during encoding/decoding while maintaining the compression performance benefits of optical flow-based prediction.
2Measurement precision
If adaptive weight factors are applied to gradient components, then predictor accuracy is improved, but processing overhead increases
Solution Approach 1:
The patent applies different weight factors to gradient components based on local characteristics of each sub-block. By adapting weights to local motion patterns and gradient magnitudes rather than using uniform weights globally, the method improves predictor accuracy for diverse motion scenarios while limiting processing overhead through localized rather than exhaustive optimization.
Solution Approach 2:
The patent dynamically adjusts weight factors based on gradient map characteristics and motion vector magnitudes. By changing weight parameters adaptively according to local image content and motion intensity, the method achieves higher predictor accuracy for complex motions while avoiding excessive processing overhead through conditional rather than always-on adaptation.
3Productivity
If pixel unit motion information is incorporated into prediction, then compression efficiency is improved, but memory requirements increase
Solution Approach 1:
The patent processes video blocks by dividing them into smaller sub-blocks and applies optical flow prediction independently to each sub-block. This segmentation enables pixel-unit motion information to be incorporated locally, improving compression efficiency through precise motion modeling while reducing memory requirements by limiting the scope of motion data that must be stored and processed simultaneously.
Data Source
AI summary
The present invention provides a method for decoding a video signal on the basis of an optical flow motion vector, comprising the steps of: when a bi-directional prediction is performed for a current block, acquiring a gradient map corresponding to the current block, wherein the gradient map represents a set of gradients at all pixel positions within a block having a larger size than the current block, and each gradient includes an X-axis gradient component and a Y-axis gradient component; determining an optical flow motion vector by applying a weight factor to the X-axis gradient component and the Y-axis gradient component; acquiring an optical flow predictor on the basis of the optical flow motion vector; and reconstructing the video signal by using the optical flow predictor.