Image Processing Subject Identification via Relative Scale Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face precision issues in identifying photographic subjects due to errors in feature point matching, particularly under noise influences, leading to decreased similarity calculation accuracy.
Innovation Solution
An image processing apparatus that detects feature points, calculates local feature values, determines correspondence points and relative scale sizes, performs clustering on feature points based on relative scale sizes, and selects feature points to compare images accurately, thereby reducing erroneous matching and enhancing identification precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If feature point matching is performed to identify photographic subjects, then subject identification capability is achieved, but matching errors occur under noise influence leading to decreased precision
Solution Approach 1:
The patent introduces relative scale size as an intermediary parameter to mediate the matching process between feature points. By calculating and comparing relative scale sizes of correspondence points, the system filters out erroneous matches caused by noise while preserving valid correspondences, thus resolving the contradiction between achieving subject identification and maintaining matching precision
Solution Approach 2:
The patent changes the parameter set used for matching by incorporating relative scale size information alongside feature point coordinates. This parameter expansion allows the system to distinguish between valid and erroneous matches more effectively, improving precision without sacrificing identification capability
2Measurement precision
If clustering is performed on all feature points to eliminate erroneous matching, then identification precision improves, but computational complexity increases
Solution Approach 1:
The patent applies local quality by performing clustering operations selectively on correspondence points with similar relative scale sizes rather than all feature points uniformly. This localized approach to clustering reduces computational complexity while maintaining the precision benefits of error elimination
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image processing apparatus includes a first local amount generation unit, a correspondence point calculation unit, a relative correspondence point information calculation unit, a correspondence point selecting unit and a decision unit. The first local feature value generation unit calculates, for a first image, a set of information about first local feature values including a first feature point(s). The correspondence point calculation unit calculates, for a second image, as information about correspondence point(s), a correspondence relation between the first feature point and a second feature point(s) contained in a set of information about second local feature values calculated from the second image. The relative correspondence point information calculation unit calculates relationships of scales of feature points, as the information about the relative scale sizes of correspondence points, based on the set of information about the first local feature values and the set of information about the second local feature values on the information about correspondence point(s). The correspondence point selecting unit performs clustering on feature points, based on the information about the relative scale sizes of the correspondence points, and selects at least one feature point, and the decision unit decides on possible photographic subject identity, from cluster to cluster, based on the feature point(s) selected.