Video Image Zone Encoding via Candidate Grouping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding methods, such as those using MPEG and H264 standards, face inefficiencies in prediction and residual coding, particularly with the 'template matching' technique, which relies on neighborhood correlations for intra and inter prediction, leading to suboptimal performance in reducing prediction error and encoding complexity.
Innovation Solution
The method determines a set of candidate zones for predicting an image area by calculating prediction error on the zone itself, rather than its neighborhood, and groups these zones to identify a representative that minimizes reconstruction error, allowing for improved intra and inter coding modes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If template matching technique is used for prediction based on neighborhood correlations, then prediction capability is provided, but prediction accuracy is insufficient and encoding complexity remains high
Solution Approach 1:
The invention divides the image into multiple zones (e.g., 4x4 pixel blocks) and performs prediction independently on each zone using template matching. This segmentation allows the complex prediction task to be broken down into smaller, manageable units, improving overall prediction accuracy while reducing the complexity burden on any single processing step.
Solution Approach 2:
The invention applies different prediction strategies to different zones within the image. Each zone is evaluated independently and can use template matching when beneficial, while other regions may use conventional prediction methods. This local quality approach optimizes prediction accuracy in specific regions without unnecessarily increasing complexity across the entire image.
2Productivity
If conventional block-based prediction is used, then encoding standard compliance is maintained, but throughput is limited and coding efficiency is suboptimal
Solution Approach 1:
The invention segments the conventional block-based prediction process into finer granular zones (4x4 pixel blocks). This segmentation enables more flexible and efficient prediction operations that can proceed in parallel or with reduced dependency chains, thereby increasing throughput while maintaining manageable coding complexity through systematic zone processing.
Solution Approach 2:
The invention introduces dynamic adaptation in the prediction process by allowing the choice between template matching and conventional prediction methods to vary based on local image characteristics and zone properties. This dynamic approach optimizes throughput by selecting the most efficient prediction method for each zone without requiring complete redesign of the entire coding system.
3Measurement precision
If template matching is applied to predict entire images, then prediction coverage is comprehensive, but computational burden increases significantly
Solution Approach 1:
The invention segments the image into multiple smaller zones (e.g., 4x4 pixel blocks) and applies template matching only to these localized regions rather than the entire image. This segmentation dramatically reduces the computational burden by limiting the scope of correlation calculations to small, manageable patches while still achieving comprehensive prediction coverage through systematic processing of all zones.
Solution Approach 2:
The invention applies template matching selectively to local zones where it provides the most benefit, rather than uniformly across the entire image. By evaluating local characteristics and applying computation only where needed, the system reduces overall computational burden while maintaining prediction error reduction in critical regions.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for coding a sequence of video images comprising a cutting into zones of at least one current image of the sequence, characterized in that, for the coding of at least one target zone of said at least one current image, it comprises the steps of: determination (E2) of a set of candidate zones in previously coded and then decoded parts of images of the sequence, minimizing a prediction error calculated over the neighborhood of the target zone, formation (E3) of a set of groups of candidate zones in said set of candidate zones, as a function of a criterion of similarity between the candidate zones, identification (E6) of a best group in said set of groups, minimizing a reconstruction error with respect to the target zone, determination (E6) of an indication (n) representative of said identified best group; and coding (E7) of said indication (n).