Suspiciousness Degree Estimation Model Generation via Face Image Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face image processing technologies face difficulties in estimating attributes with unclear class definitions, such as suspiciousness degree, without pre-defined reference features, making it challenging to detect unregistered suspicious individuals.
Innovation Solution
A suspiciousness degree estimation model is generated through clustering face images based on extracted features and associated suspiciousness degree information, allowing for the estimation of suspiciousness without relying on pre-defined attributes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a reference feature such as an eigenvector for each age group is held to compare with extracted face features, then age estimation can be performed, but it becomes impossible to estimate attributes with unclear class definitions like suspiciousness degree
Solution Approach 1:
The patent introduces a clustering unit as an intermediary that groups face images into clusters based on extracted features without requiring pre-defined class labels. This clustering mechanism serves as a mediator between feature extraction and attribute estimation, enabling the system to handle attributes with unclear class definitions by discovering natural groupings in the data rather than relying on pre-established categories
Solution Approach 2:
The patent changes the approach from using fixed reference features with predefined classes to using dynamic clustering results that adapt to the data distribution. By transforming the estimation process into a clustering-based approach, the system can flexibly accommodate various attributes including suspiciousness degree without being constrained by rigid class definitions
2Adaptability or versatility
If face images are clustered based on extracted features without pre-defined class labels, then attributes with unclear class definitions can be estimated, but the system complexity increases
Solution Approach 1:
The patent segments the face image database into multiple clusters based on extracted features, allowing the system to handle complex attribute estimation by breaking down the data into manageable groups. This segmentation enables the clustering unit to process large datasets efficiently by organizing them into distinct categories that can be independently analyzed for attribute estimation
3Reliability
If pre-defined reference features are used for attribute estimation, then estimation can be performed for attributes with clear class definitions, but detection of unregistered suspicious individuals becomes difficult
Solution Approach 1:
The clustering unit performs self-organization of face images into clusters based on their intrinsic features without requiring external reference data or pre-defined categories. This self-service mechanism enables the system to automatically discover patterns and groupings that can identify suspicious individuals without relying on pre-registered blacklisted persons, thereby improving detection capability for unregistered suspects
Data Source
AI summary
A suspiciousness degree estimation model generation device includes: a clustering unit that performs clustering on an input face image based on the feature extracted from the face image; and a suspiciousness degree estimation model generation unit that generates a suspiciousness degree estimation model used for estimating the suspiciousness degree of an estimation target person, based on the result of clustering by the clustering unit and suspiciousness degree information that is previously associated with a face image included by the clustering result and that shows the suspiciousness degree of a person shown by the face image. The suspiciousness degree estimation device includes: a feature extraction unit that extracts a feature from a face area of an estimation target person; and a suspiciousness degree estimation unit estimates the suspiciousness degree of the estimation target person, based on the feature extracted by the feature extraction unit and the suspiciousness degree estimation model generated by the suspiciousness degree estimation model generation device.