Pattern Matching Pyramid with Dynamic Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In image processing for pattern matching, existing methods face challenges in maintaining detection accuracy while increasing processing speed, particularly in industrial applications, where reducing image resolution for pyramid searching can lead to unreliable similarity calculations and potential object detection failures.
Innovation Solution
The method generates a target object pyramid with varying reduction ratios and performs pattern matching by scanning with a model image in each layer, extracting candidate points with a degree of similarity above a threshold, and dynamically updating an area threshold for further similarity calculations, allowing for early termination of calculations when similarity falls below a certain value.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If pyramid searching is performed with reduced resolution images to increase processing speed, then productivity is improved, but reliability deteriorates due to inaccurate similarity calculations
Solution Approach 1:
The patent divides the image processing into multiple pyramid layers with different reduction ratios. Pattern matching is performed sequentially from lower layers (higher resolution) to upper layers (lower resolution). This segmentation allows accurate detection at lower layers while using upper layers for efficient candidate filtering, thus resolving the contradiction between processing speed and detection accuracy.
Solution Approach 2:
The patent applies different processing strategies to different pyramid layers. Lower layers use full pattern matching for accuracy, while upper layers use simplified matching with early termination for speed. This local quality differentiation optimizes the balance between reliability and productivity at each level of the pyramid structure.
2Reliability
If pattern matching is performed at every position in the target object image to ensure accurate detection, then reliability is improved, but productivity deteriorates due to high calculation cost
Solution Approach 1:
The patent performs preliminary pattern matching at lower pyramid layers before proceeding to higher layers. This preliminary action identifies candidate regions that are likely to contain the target object, allowing the system to skip exhaustive matching in non-candidate regions at higher layers, thus improving productivity while maintaining reliability.
Solution Approach 2:
The patent implements early termination of similarity calculations when the degree of similarity falls below a threshold. This skipping mechanism avoids unnecessary calculations in regions that cannot contain the target object, significantly improving processing speed while maintaining detection accuracy through the multi-layer pyramid structure.
3Productivity
If early termination of similarity calculation is implemented to increase processing speed, then productivity is improved, but reliability deteriorates due to potential false negatives
Solution Approach 1:
The patent uses lower pyramid layers as a cushioning mechanism before applying early termination at higher layers. The lower layers perform complete pattern matching to establish reliable candidate regions, which then serve as a foundation for the early termination strategy at higher layers. This beforehand cushioning prevents false negatives while enabling speed optimization.
Solution Approach 2:
The patent uses feedback from lower pyramid layers to guide the matching process at higher layers. Candidate regions identified at lower layers inform the search space and threshold settings at higher layers, ensuring that early termination does not lead to false negatives. The feedback mechanism maintains reliability while enabling productivity improvements.
Data Source
AI summary
To each of points at a position corresponding to each of candidate points extracted from a target object image in a layer upper than a target object image, an area including the point at the position corresponding to the candidate point and an area threshold are set. As a result of predicting a degree of similarity, in a case where the degree of similarity has a value smaller than the area threshold, calculation of the degree of similarity is terminated in the middle. In a case where the degree of similarity has a value equal to or greater than the area threshold, calculation of the degree of similarity is continued, and in a case where the degree of similarity has a value greater than the area threshold when calculation of the degree of similarity is completed, a next area threshold is updated with a value greater than the area threshold.