Face Authentication Clustering Display Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face authentication techniques using deep learning often misclassify face images due to low similarity between clusters with high and low likelihood, leading to errors in annotation and correction processes.
Innovation Solution
An information processing apparatus that includes a clustering unit, a determination unit, and selection unit to identify and display face images from clusters with varying degrees of similarity, ensuring accurate representation of face images by selecting images from the core cluster, the first cluster with the lowest similarity, and the second cluster with intermediate similarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If clusters are selected sequentially from highest to lowest likelihood for display, then the most representative cluster is displayed first, but face images from dissimilar clusters are displayed together causing misclassification errors
Solution Approach 1:
The patent segments the display process into three distinct phases: first displaying the core cluster with highest likelihood, then displaying clusters with intermediate similarity degrees, and finally displaying clusters with lowest similarity. This segmentation prevents mixing of dissimilar face images while maintaining systematic progression through the clustering results.
Solution Approach 2:
The patent implements periodic display actions where face images are presented in multiple rounds or phases. In each period, the system displays a specific subset of clusters based on their similarity degree to the core cluster, allowing operators to review and correct annotations in organized batches rather than all at once.
2Loss of information
If all clusters are displayed together on the screen, then complete clustering results are visible, but face images from different human figures are mixed making correction difficult
Solution Approach 1:
The patent divides the complete set of clustering results into multiple segments based on similarity degree. Each segment contains face images from clusters with comparable similarity characteristics, allowing operators to focus on one segment at a time while maintaining awareness of the complete dataset through systematic navigation between segments.
Solution Approach 2:
The patent implements dynamic display control where the system adaptively adjusts which clusters are displayed based on operator interaction. The display transitions dynamically between different similarity degree groupings, and the system can reorganize displayed content based on correction feedback, making the interface responsive to operational needs.
Data Source
AI summary
An apparatus includes a clustering unit configured to perform clustering on a data group based on a feature value of each of a plurality of pieces of data, a determination unit determines a representative cluster among clusters generated by the clustering unit, a first identification identifies a first cluster based on a degree of similarity between each of the clusters generated by the clustering unit and the representative cluster, a second identification unit identifies a second cluster based on a first degree of similarity, which is the degree of similarity between each of the clusters generated by the clustering unit and the representative cluster, and a second degree of similarity, which is a degree of similarity between each of the clusters generated by the clustering unit and the first cluster, and a selection unit selects data for display from among the representative cluster, the first cluster, and the second cluster.