Localized Simple Multiple Kernel K-Means Clustering for Face Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing SimpleMKKM algorithm tightly aligns the combined kernel matrix with globally generated ideal similarity, neglecting local structures and resulting in unsatisfactory clustering performance for face image data.
Innovation Solution
A localized simple multiple kernel k-means clustering method is introduced, which calculates localized kernel matrices using nearest neighbor matrices and constructs an optimized objective function using a reduced gradient descent method to improve clustering partition matrices, enhancing local information representation and fusion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If SimpleMKKM tightly aligns the combined kernel matrix with globally generated ideal similarity, then the optimization framework is simplified, but local structures are neglected and clustering performance deteriorates
Solution Approach 1:
The patent segments the global alignment problem into local alignment problems by introducing localized kernel matrices. Instead of aligning the entire combined kernel matrix with the ideal similarity matrix globally, the method divides the alignment into local neighborhoods using k-nearest neighbor graphs. Each sample is aligned with its local neighbors, preserving local structures while maintaining computational feasibility.
Solution Approach 2:
The patent applies local quality by making the alignment process adaptive to local data characteristics. The localized kernel matrix for each sample is constructed based on its specific neighborhood, allowing different regions of the data to be aligned with appropriate local similarities rather than forcing a uniform global alignment. This ensures that local structures are preserved and utilized in the clustering process.
2Device complexity
If global alignment is used to simplify optimization, then computational complexity is reduced, but the representability of local structures is lost
Solution Approach 1:
The global alignment task is segmented into multiple local alignment tasks. The patent constructs localized kernel matrices by considering only k-nearest neighbors for each sample, thereby segmenting the overall alignment problem into manageable local components. This segmentation preserves local structure information while avoiding the computational burden of global alignment.
Solution Approach 2:
The patent introduces k-nearest neighbor graphs as an intermediary structure to facilitate local alignment. These graphs serve as mediators that connect samples to their local neighbors, enabling the alignment process to focus on local structures. The intermediary neighbor graphs preserve local topology information that would otherwise be lost in direct global alignment approaches.
3Measurement precision
If localized kernel matrices are used to preserve local structures, then clustering accuracy improves, but computational complexity increases
Solution Approach 1:
The computation is segmented by processing each sample's local neighborhood independently rather than computing global alignments. The localized kernel matrix for each sample is computed only using its k-nearest neighbors, which reduces the effective computation scope. Although more computations are needed for localization, each individual computation is simpler and more focused.
Solution Approach 2:
The patent applies partial action by computing localized kernel matrices only for necessary local neighborhoods rather than performing exhaustive global computations. The k-nearest neighbor approach computes only the necessary local structures required for accurate clustering, avoiding excessive computation on distant or irrelevant samples while still achieving high clustering accuracy.
Data Source
AI summary
A face image clustering method and system based on a localized simple multiple kernel k-means is provided. The face image clustering method based on localized simple multiple kernel k-means includes the following steps: S1, acquiring face images, and preprocessing the acquired face images to obtain an average kernel matrix for each view; S2, calculating n (Σ×n)-nearest neighbor matrices according to the obtained average kernel matrices; S3, calculating a localized kernel matrix for each view according to the nearest neighbor matrices; S4, constructing a localized simple multiple kernel k-means clustering objective function according to the calculated localized kernel matrix for each view; S5, solving a minimum of the constructed objective function by adopting a reduced gradient descent method to obtain an optimal clustering partition matrix; and S6, performing k-means clustering on the obtained clustering partition matrix to achieve clustering.