Image Matching System Using SVM Classifier for False Positive Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing techniques for object recognition and identification in images lack scalability and accuracy for real-time, on-the-fly processing of multiple images, leading to high false positive matches and reduced precision and recall.
Innovation Solution
An image processing system that utilizes a combination of feature extraction methods such as SIFT, SURF, and GLOH, along with a support vector machine classifier, to filter and classify candidate matches, ensuring accurate identification of true positive matches by employing geometric constraints and a weighted scoring system to improve precision and recall.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image matching techniques are used for real-time processing of multiple images, then processing speed is maintained, but accuracy and precision deteriorate due to high false positive matches
Solution Approach 1:
The patent segments the image matching process into multiple stages: initial feature matching, candidate generation, and classifier-based verification. This multi-stage segmentation allows traditional fast matching to be combined with more accurate but computationally intensive classification only on candidate matches, resolving the contradiction between speed and accuracy
Solution Approach 2:
The system performs preliminary feature extraction and candidate match generation before applying the full classifier. By pre-processing images to identify only promising candidates, the system reduces the computational burden of accurate classification while maintaining real-time processing capability
2Reliability
If traditional image matching techniques are used without filtering, then processing speed is maintained, but reliability deteriorates due to high false positive matches
Solution Approach 1:
The patent introduces an intermediary classifier system that acts as a mediator between simple feature matching and final match confirmation. This classifier uses geometric constraints and scoring mechanisms to filter false positives without requiring complete re-analysis of all matches, balancing reliability improvement with controlled complexity
Solution Approach 2:
The system changes parameters dynamically by adjusting scoring thresholds and geometric constraint strictness based on the confidence level of initial matches. High-confidence matches undergo simpler verification, while low-confidence matches receive more rigorous filtering, optimizing the balance between reliability and complexity
3Measurement precision
If comprehensive feature extraction methods are used, then measurement precision is improved, but loss of time increases due to processing overhead
Solution Approach 1:
The patent applies partial action by using comprehensive feature extraction only on candidate regions identified by faster preliminary methods. Instead of applying full SIFT, SURF, and GLOH extraction to all images, the system applies these methods selectively to regions likely to contain matches, reducing processing time while maintaining precision
Data Source
AI summary
An image matching technique locates feature points in a template image such as a logo and then does the same in a test image. Classifiers are trained for multiple template images and the classifiers are used to evaluate a match between a template image and a test image.