Feature Descriptor Selection for Mobile Image Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for feature descriptor matching in images are inefficient on mobile devices due to high memory requirements and computational load, especially when dealing with large numbers of reference descriptors and varying conditions such as viewpoint, illumination, and camera parameters.
Innovation Solution
A method that creates a reduced set of representative feature descriptors by aggregating descriptors from multiple views of an object, selecting the most significant ones based on a scoring system, and storing these in a database for efficient matching and localization, which can be applied on mobile devices with limited memory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large number of reference feature descriptors are used to improve matching accuracy under varying conditions, then the success rate of feature matching increases, but the memory requirements and computational load increase making real-time processing impossible
Solution Approach 1:
The patent extracts and stores only the most representative and discriminative feature descriptors from a large set of synthetic views. By selecting key descriptors that capture essential object characteristics across different conditions, the system achieves high matching accuracy with a reduced descriptor set, resolving the contradiction between matching reliability and memory consumption.
Solution Approach 2:
The patent performs preliminary processing offline to generate and evaluate multiple synthetic views, pre-selecting the most effective descriptors before deployment. This preliminary action allows the system to identify and store only the essential descriptors needed for accurate matching, eliminating the need to store and process all possible descriptors during real-time operation.
2Adaptability or versatility
If multiple synthetic views are created to simulate different appearances and improve invariance to viewpoint variations, then the robustness of feature matching increases, but the amount of reference data increases resulting in longer loading times
Solution Approach 1:
The patent extracts only the most representative descriptors from multiple synthetic views rather than storing all synthetic view data. By selecting key descriptors that capture essential object characteristics across different viewpoints, the system achieves robust invariance to viewpoint variations while minimizing the amount of data that needs to be loaded and stored.
Solution Approach 2:
The patent creates multiple synthetic views to ensure adequate coverage of viewpoint variations, but then selects only the essential subset needed for robust matching. This partial action approach provides sufficient invariance without the excessive data storage and loading time that would result from using all synthetic views.
3Measurement precision
If feature classifiers are trained with randomized trees and probabilities stored for every leaf node to improve classification accuracy, then the precision of feature identification increases, but the memory requirements become prohibitive for mobile devices
Solution Approach 1:
The patent extracts and stores only the most significant descriptors identified through the scoring mechanism rather than maintaining complete probability distributions for all classes in randomized trees. This extraction of essential information maintains classification precision while dramatically reducing memory requirements to levels suitable for mobile devices.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of providing a set of feature descriptors comprises: a) providing images of a first object or images of multiple different appearances or different versions of a first object, b) extracting at least one feature from these images, c) providing descriptors for the extracted features and storing the descriptors in a first set of descriptors, d) matching a plurality of the descriptors of the first set of descriptors, e) computing a score parameter for a plurality of the descriptors based on the result of the matching process, f) selecting at least one descriptor based on its score parameter in comparison with score parameters of other descriptors, g) adding the selected descriptor to a second set of descriptors, h) updating the score parameter of a plurality of the descriptors according to any preceding selection process and to the result of the matching process, i) performing steps, f) and g) again.