Image Encoding Template Matching Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image encoding and decoding technologies face high calculation complexities and errors due to the reliance on distortion between templates and matching target regions, leading to inefficient prediction block determination.
Innovation Solution
The method involves determining a matching search target image using encoding information, neighboring block information, and user-defined parameters to find an optimal prediction block by calculating distortion and similarity, and generating residual blocks through filtering and offset subtraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If template matching is performed using all restored reference images, then prediction accuracy is improved, but calculation complexity increases
Solution Approach 1:
The patent divides the template matching process into two distinct stages: coarse matching and fine matching. The coarse matching stage uses a simplified distortion calculation to quickly identify candidate regions, while the fine matching stage applies more accurate similarity metrics only to these candidates. This segmentation reduces the overall calculation complexity while maintaining prediction accuracy.
Solution Approach 2:
The patent applies partial action by performing comprehensive template matching only on selected candidate regions rather than all restored reference images. The coarse matching stage performs a preliminary search to identify promising areas, and the fine matching stage then focuses computational resources only on these partial regions, avoiding unnecessary calculations on irrelevant areas.
2Device complexity
If only distortion between template and matching target region is used, then calculation complexity is reduced, but prediction accuracy deteriorates
Solution Approach 1:
The patent segments the matching evaluation into two parts: distortion calculation for coarse filtering and similarity calculation for fine evaluation. The distortion metric (such as SAD or SSD) is used in the first stage to quickly eliminate poor candidates, while the similarity metric (such as correlation or MSE) is applied in the second stage to accurately assess the quality of candidate prediction blocks.
Solution Approach 2:
The patent introduces an intermediary approach by using distortion as a preliminary filter before applying similarity metrics. This intermediary distortion calculation acts as a gateway that reduces the search space, allowing the more computationally intensive similarity calculations to be performed only on a reduced set of candidates, thus balancing complexity and accuracy.
3Measurement precision
If multiple reference images are searched, then prediction block quality is improved, but processing time increases
Solution Approach 1:
The patent implements periodic action through its two-stage matching process. The coarse matching phase periodically screens multiple reference images using fast distortion calculations to identify candidate regions, and the fine matching phase periodically evaluates these candidates using accurate similarity metrics. This periodic two-phase approach efficiently processes multiple reference images without requiring continuous intensive computation across all images.
Solution Approach 2:
The patent applies preliminary action by performing coarse distortion-based matching on multiple reference images before conducting fine similarity-based evaluation. This preliminary coarse matching stage pre-identifies candidate regions across multiple reference images, allowing the subsequent fine matching to focus only on these pre-selected candidates, thereby reducing overall processing time while maintaining quality.
Data Source
AI summary
Provided are a method and an apparatus for encoding images using template matching and a method and an apparatus for decoding images. The method for encoding the images can determine a template of an encoding target block and can determine a matching search target image for performing a matching search with the determined template among the recovered reference images. In addition, an optimum prediction block can be determined on the basis of the determined matching search target image and the template of the encoding target block.