SIFT Descriptor Thresholding for GPS-Denied Navigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image matching technologies face inefficiencies in detecting similar features across pixilated images, particularly under varying illumination conditions and non-linear contrast changes, with arbitrary thresholding methods limiting performance in GPS-denied environments.
Innovation Solution
The implementation of meaningful clamping (MC) for scale invariant feature transform (SIFT) descriptors, which computes a unique bin magnitude threshold using a contrario methodology to determine perceptible deviations from uniform noise, improving matching performance by automatically adjusting the clamping threshold for each descriptor.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If arbitrary thresholding methods are used for SIFT descriptors, then the device complexity is reduced, but the measurement precision and matching accuracy deteriorate
Solution Approach 1:
The patent combines multiple descriptor statistics (mean, standard deviation, skewness, kurtosis) into a unified thresholding framework that adaptively determines clamping thresholds. This merging of statistical measures enables accurate differentiation of meaningful features from noise without requiring complex manual threshold selection procedures.
Solution Approach 2:
The thresholding method performs self-adjustment by automatically computing clamping thresholds based on the statistical properties of each descriptor bin. The system serves itself by using its own descriptor statistics to determine appropriate thresholds, eliminating the need for external arbitrary parameter specification while maintaining high matching accuracy.
2Speed
If fixed clamping threshold is applied to all descriptors, then the processing speed is improved, but the adaptability to varying illumination conditions deteriorates
Solution Approach 1:
The patent applies different clamping thresholds to different descriptor bins based on their local statistical characteristics. Each bin receives a threshold adapted to its specific mean, standard deviation, and higher-order moments, allowing the system to optimize for local feature properties rather than applying a uniform global threshold. This local adaptation maintains processing efficiency while significantly improving performance under varying illumination.
Solution Approach 2:
The clamping threshold becomes a dynamic value that adapts to each descriptor's statistical properties rather than remaining fixed. The threshold adjusts automatically based on the descriptor's mean, variance, and higher-order moments, enabling the system to respond to changing illumination conditions while maintaining consistent processing speeds through efficient statistical computation.
3Loss of information
If no thresholding is applied to descriptors, then the loss of information is minimized, but the reliability of feature detection under noise deteriorates
Solution Approach 1:
The patent changes the parameter threshold dynamically based on descriptor statistics rather than using a fixed value. By computing thresholds from the mean, standard deviation, skewness, and kurtosis of each bin, the system adapts the thresholding parameter to match the local characteristics of the data, thereby reliably distinguishing signal from noise while preserving meaningful descriptor information.
Data Source
AI summary
Embodiments are directed to image matching using local image descriptors thresholding. An image matching tool is associated with at least one electronic processor. The image matching tool is configured to determine a unique bin magnitude threshold descriptor for a test image and an image of a known object of interest stored in a database. The image matching tool determines a classification match of the test image to the image of a known object of interest.