Image Clustering Precision via Two-Stage Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face clustering in computer vision is challenged by variations in orientation, illumination, expression, occlusion, and definition of face images, leading to reduced clustering precision.
Innovation Solution
An image clustering method that obtains face and attribute features, performs initial clustering based on face features, determines similarity matrices, and re-clusters using graph convolution processing to optimize the classification of images into more precise groups.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If face clustering is performed using traditional methods, then the process is simple, but clustering precision deteriorates due to variations in orientation, illumination, expression, occlusion, and definition
Solution Approach 1:
The patent segments the clustering process into two distinct stages: initial clustering based on face features to obtain first class groups, and re-clustering based on similarity matrices to obtain second class groups. This segmentation allows each stage to focus on specific aspects of the problem, improving overall precision while managing complexity through structured division of the clustering task.
Solution Approach 2:
The patent performs preliminary clustering using face features before conducting the final re-clustering. This preliminary action creates initial groupings that serve as a foundation for the subsequent similarity matrix-based re-clustering, allowing the system to progressively refine results and achieve higher precision through staged processing.
2Measurement precision
If re-clustering based on similarity matrices is performed, then clustering precision is improved, but computational complexity increases
Solution Approach 1:
The patent divides the computational workload into two phases: initial clustering that creates first class groups with moderate computational requirements, and re-clustering that processes smaller subsets using similarity matrices. This segmentation reduces peak computational power needs while maintaining high precision through the second stage refinement.
Solution Approach 2:
The patent applies re-clustering selectively to images within first class groups rather than processing the entire image set uniformly. This partial action approach focuses computational resources on refining specific groups where precision improvements are most needed, rather than performing exhaustive processing on all data.
Data Source
AI summary
An image clustering method and apparatus, an electronic device, and a storage medium are provided. The method includes: obtaining a face feature and an attribute feature of each image in an image set; performing clustering processing on the image set by using a similarity of the face feature of each image in the image set to obtain at least one first class group, images included in each first class group being predicted images including a same object; determining a similarity matrix for each first class group by using the face features and the attribute features of the images in each first class group; and re-clustering the images in each first class group based on the similarity matrix corresponding to each first class group and the face features of the images in each first class group to obtain at least one second class group, images in the second class group being optimized predicted images having a same object.