Encoder and Decoder Optical Flow Prediction with Reduced Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The computation-intensive process of deriving parameters for bi-directional optical flow in video coding increases the processing load, hindering efficient video encoding and decoding.
Innovation Solution
An encoder and decoder that derive parameters for generating prediction images using sums of horizontal and vertical gradient values, reducing computation by inverting or maintaining the sign of pixel difference values based on gradient sums, and using these parameters to generate prediction images with reduced processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If bi-directional optical flow parameters are derived using conventional methods, then prediction accuracy is improved, but computation load increases
Solution Approach 1:
The patent segments the bi-directional optical flow parameter derivation into multiple components: gradient calculation at reference pixel positions, pixel difference calculation, and sign inversion operations. This segmentation allows the computation to be distributed and optimized, reducing the overall computation load while maintaining prediction accuracy through systematic processing of gradient and difference values.
Solution Approach 2:
The patent performs preliminary calculations of gradient values and pixel differences at reference block positions before generating the final prediction image. By pre-computing these parameters and storing them, the system avoids redundant calculations during the actual prediction process, thereby reducing computation load while preserving accuracy.
2Measurement precision
If detailed gradient calculations are performed for each pixel position, then prediction precision is improved, but processing time increases
Solution Approach 1:
The patent extracts only the essential gradient information and pixel difference values at specific reference positions rather than performing exhaustive calculations at every pixel position. This extraction approach focuses computational effort on the most critical parameters needed for accurate prediction, reducing processing time while maintaining precision.
Solution Approach 2:
The patent changes the computational parameters by working with gradient sums and signed difference values rather than raw pixel intensities. This parameter transformation simplifies the mathematical operations required, enabling faster computation while preserving the precision needed for accurate bi-directional optical flow prediction.
Data Source
AI summary
An encoder includes circuitry and memory connected to the circuitry. The circuitry: derives an absolute value of a sum of horizontal gradient values; derives, as a first parameter, the total sum of the absolute values of horizontal gradient values; derives, as a second parameter, the total sum of the absolute values of vertical gradient values; derives a horizontal-related pixel difference value; derives, as a third parameter, the total sum of the absolute values of horizontal-related pixel difference values; derives a vertical-related pixel difference value; derives, as a fourth parameter, the total sum of the absolute values of vertical-related pixel difference values; and generates a prediction image using the first to fourth parameters.