Image Prediction Encoding Device Reducing Computational Time
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image prediction methods suffer from distortion in prediction signals due to encoding, leading to decreased encoding efficiency and image quality, particularly when dealing with images with large motion, as they require wider search ranges and increased computational time.
Innovation Solution
An image prediction encoding device and method that generates multiple candidate prediction signals using reduced motion information within a shorter search time, allowing for efficient production of averaged prediction signals with limited additional information and computational resources, while also limiting the search range.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a wider search range is used to handle images with large motion, then prediction accuracy is improved, but computational time and complexity increase
Solution Approach 1:
The patent segments the search process into multiple stages: first generating candidate motion vectors using reduced motion information for quick filtering, then performing full search only on promising candidates. This segmentation allows handling large motion effectively while reducing overall computational time by avoiding exhaustive search across the entire wide search range.
Solution Approach 2:
The patent performs preliminary action by generating candidate prediction signals using reduced motion information before the main prediction process. This preliminary step identifies promising candidates that can then be evaluated in full, reducing the need to search the entire wide range and thus decreasing computational time while maintaining prediction accuracy for large motion.
2Measurement precision
If multiple candidate prediction signals are generated and averaged, then prediction quality is improved by reducing noise, but encoding complexity increases
Solution Approach 1:
The patent applies partial action by generating a limited number of candidate prediction signals (not all possible candidates) and averaging only those. This selective averaging reduces noise in prediction signals while keeping encoding complexity manageable by processing only a subset of candidates rather than all possible predictions.
Solution Approach 2:
The patent merges multiple candidate prediction signals through averaging to produce a final prediction signal. This combining process reduces noise components while improving prediction quality, and the method manages complexity by efficiently selecting and merging only the most relevant candidates.
3Productivity
If motion information is reduced for faster processing, then encoding speed is improved, but prediction accuracy may deteriorate
Solution Approach 1:
The patent uses reduced motion information as a preliminary step to quickly generate candidate prediction signals, improving encoding speed. Then, these candidates are refined and evaluated to ensure prediction accuracy is maintained, thus resolving the trade-off between speed and precision.
Solution Approach 2:
The patent substitutes the traditional mechanical exhaustive search with a more efficient system that uses reduced motion information to guide the search. This substitution replaces brute-force computation with a smarter, information-guided approach that maintains accuracy while improving encoding speed.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention is directed to efficient production of a prediction signal with small additional information and small computational amount. An image prediction encoding device 100 includes a prediction signal generator 103 that produces a prediction signal with respect to a target pixel signal of a target region that is among a plurality of regions divided by a block division unit 102, a motion estimator 113 that searches for a motion vector required for the production of the prediction signal from an already reproduced signal, and a subtractor 105 that produces a residual signal indicating a difference between the prediction signal and the target pixel signal. The prediction signal generator 103 determines a search region that is a partial region of the already reproduced signal based on the motion vector and produces one or more type 2 candidate prediction signals. The prediction signal generator 103 includes a candidate prediction signal combining unit 203 that produces the prediction signal by processing the plurality of type 2 candidate prediction signals and a type 1 candidate prediction signal produced based on the motion vector in accordance with a predetermined combining method.