Keypoint Matching via Eigenvalue Analysis for Image Comparison

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image comparison methods are time-consuming and require significant data processing, especially when dealing with large reference databases, which hampers the efficiency of object recognition processes.

Innovation Solution

A method that identifies correct keypoint matches by using the dominant eigenvalue and outlier probability density functions to distinguish inliers from outliers, allowing for efficient image comparison by determining the number of correct keypoint matches and identifying which matches are most likely correct.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the number of reference images in the database is increased to improve object recognition accuracy, then the reliability of object recognition is improved, but the time required for comparing operations and data processing requirements increase significantly

Engineering Contradiction:
Improveobject recognition accuracyVSAvoidcomparing operation time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent segments the reference database into multiple subsets or groups, allowing the system to perform comparing operations on smaller portions of data at a time. This segmentation reduces the computational burden for each individual comparison while maintaining comprehensive coverage across all reference images, thereby resolving the contradiction between database size and processing time.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary processing and organization of reference images before the actual object recognition process. By pre-computing features, organizing images into categories, or creating index structures in advance, the system reduces the time required for during-comparison operations, allowing large databases to be searched efficiently without linearly increasing comparison time.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If the number of reference images is increased to improve object recognition reliability, then the reliability is improved, but the data processing requirements increase

Engineering Contradiction:
Improveobject recognition accuracyVSAvoiddata processing efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent extracts and utilizes only the essential features or characteristics from reference images that are most relevant for object recognition. By focusing on key discriminative features rather than processing complete image data, the system maintains high recognition accuracy with reduced data processing requirements, effectively resolving the contradiction between database comprehensiveness and processing efficiency.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the parameter representation of reference images from full-resolution images to compressed feature vectors, extracted descriptors, or other parameterized forms. This parameter transformation reduces the data volume and processing complexity while preserving the essential information needed for accurate object recognition, thereby improving data processing efficiency without sacrificing reliability.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If traditional keypoint matching procedures are used to distinguish inliers from outliers, then the measurement precision of keypoint matches is improved, but the time consuming nature of the procedure increases

Engineering Contradiction:
Improvekeypoint match accuracyVSAvoidinlier identification time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies partial verification procedures to keypoint matches, checking only the most critical criteria or using a two-stage approach where a fast preliminary filter identifies likely inliers followed by more rigorous verification only for candidates that pass the initial screen. This partial action maintains measurement precision for final results while significantly reducing the time spent on exhaustive verification of all matches.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The patent replaces traditional iterative mechanical procedures for inlier identification with more efficient computational methods such as randomized algorithms, parallel processing approaches, or hardware-accelerated computations. This substitution maintains the precision of inlier identification while dramatically reducing the time required by leveraging different computational paradigms or physical implementations.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP2801054B1Method and system for comparing images
Publication Date: 2017.06.28 TELECOM ITALIA SPA
  • EP2801054B1 patent drawingFigure 1A
  • EP2801054B1 patent drawingFigure 1B
  • EP2801054B1 patent drawingFigure 1C

AI summary

A method for comparing a first image with a second image, comprises: identifying first keypoints in the first image and second keypoints in the second image; forming matches between first and second keypoints by associating each first keypoint with a corresponding second keypoint; calculating a plurality of first distance ratios, each first distance ratio being based on the ratio of the distance between a pair of said first keypoints and the distance between a corresponding pair of said second keypoints matching the pair of first keypoints; arranging a distribution of values of said calculated plurality of first distance ratios in a histogram having a plurality of ordered bins each corresponding to a respective interval of values of the first distance ratios, the histogram enumerating for each bin a corresponding number of first distance ratios of the distribution having values comprised within the respective interval; generating an outlier density function expressing a statistical distribution of a plurality of second distance ratios corresponding to a random selection of keypoints pairs in the first and second images, said plurality of second distance ratios being based on the ratio of the distance between pairs of first keypoints in the first image and the distance between corresponding pairs of randomly selected second keypoints in the second image; discretizing said outlier density function by generating for each of said bin a corresponding outlier probability density value. The method further comprises: determining a number of correct keypoint matches, each formed by a first and a second keypoints corresponding to a same point of a same object depicted in the first and second images, said determining a number of correct keypoint matches including: determining a matrix, each element of said matrix corresponding to a respective pair of keypoint matches and having a value corresponding to the difference between the value of said histogram at the bin including the distance ratio of the respective pair of keypoint matches and the outlier probability density value corresponding to said bin, said outlier probability density value being weighted by means of a parameter (beta); determining a value of the parameter (beta) such that the dominant eigenvector of the matrix is equal to a vector (r) having elements of a first value if the pair of keypoint match is a correct keypoint match an of a second value if the pair of keypoint match is an incorrect keypoint match; determining the number of correct keypoint matches as an expression of the dominant eigenvalue associated to said dominant eigenvector, wherein for comparing said first image with said second image the determined number of correct keypoint matches is exploited.