Image Processing Device Feature Point Matching Partial Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing devices face challenges in accurately identifying objects across images with changes in shape, as they struggle to distinguish between correct and incorrect feature point pairs, leading to decreased accuracy in geometric transformation modeling and object identification.
Innovation Solution
An image processing device that detects feature points, calculates local feature amounts, identifies correlations between feature points in different images, and determines object similarity by analyzing partial regions within the images, using methods such as partial region detection and geometric validation to improve accuracy even when objects have shape changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If geometric validation is performed on all corresponding points using RANSAC, then the geometric transformation model can be estimated, but the accuracy decreases when objects have shape changes because incorrect feature point pairs cannot be distinguished
Solution Approach 1:
The patent segments the image into multiple analysis regions and performs feature point matching and geometric validation independently in each region. This segmentation allows the system to handle local shape variations within each region while maintaining overall accuracy, resolving the contradiction by avoiding the need to process all corresponding points globally with RANSAC.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. By performing geometric validation locally in each analysis region rather than globally, the system can maintain high accuracy in regions with consistent geometry while tolerating shape changes in other regions, thus resolving the accuracy-reliability contradiction.
2Area of stationary object
If all feature points are used for matching, then the coverage of the object is maximized, but the presence of incorrect feature point pairs reduces the accuracy of object identification
Solution Approach 1:
The patent divides the image into multiple analysis regions and performs feature point matching in each region separately. This segmentation strategy maintains comprehensive coverage by processing all regions while improving accuracy by limiting each matching operation to a localized area where shape variations are minimized, thus resolving the coverage-accuracy contradiction.
Solution Approach 2:
The patent performs geometric validation on a subset of corresponding points within each analysis region rather than all feature points globally. This partial action approach maintains sufficient coverage through multiple regions while improving accuracy by validating only the most reliable local correspondences in each region.
Data Source
AI summary
Provided is an image processing device that can suppress deterioration in the accuracy of identifying a subject, even in cases where the shape of the subject in an image is deformed.A feature amount calculation means 81 detects feature points from an image, and calculates, for each feature point, a local feature amount for the feature point on the basis of a peripheral region of said feature point, including said feature point. A correlation identifying means 82 specifies the correlation between feature points in a first image and feature points in a second image on the basis of the local feature amount of each feature point in the first image and the local feature amount of each feature point in the second image. A partial region detection means 83 detects, from one of the first image or the second image, partial regions each including a feature point in said image. A matching means 84 determines, for each partial region, whether or not a subject in the first image is identical with or similar to a subject in the second image on the basis of the feature point included in the partial region and a feature point corresponding to the feature point.