Multi-view atlas clustering method and system based on graph optimization

Abstract

The invention discloses a multi-view graph clustering method and system based on graph optimization, and the method comprises the steps: carrying out the clustering of a multi-view data set through an alternating iteration optimization method, and generating a fusion graph; based on the data set, a data similarity matrix is constructed, and the data similarity matrix is used for obtaining the similarity between the feature points in each view; designing a target function and a constraint condition for clustering based on the data similarity matrix; based on an alternating iterative optimization method, performing iterative processing on the target function until the target function converges or reaches the maximum number of iterations; and obtaining the data category indication vector after iteration processing, and generating a fusion image. According to the method, an alternating iteration method is adopted to generate stable clusters, randomness is avoided, and compared with a traditional data division technology, the accuracy is improved by at least 10%.

Description

technical field [0001] The invention relates to the field of computer technology, and in particular, to a multi-view spectral clustering method and system based on graph optimization. Background technique [0002] With the rapid development of information extraction technology, multi-view data has been widely used in many fields. For example, a sentence can be spoken in different voices; an article can be translated into several languages; an object can be viewed from different perspectives, and so on. The heterogeneity of multi-view data brings challenges to its clustering. Among the multi-view clustering methods, graph theory-based methods have attracted much attention. The purpose of this method is to cut the constructed similarity graph by minimizing the inter-cluster similarity and maximizing the intra-cluster similarity. Similarity graphs can be learned from different views, and cutting methods are usually done by performing spectral clustering (often consisting of ...

AI Technical Summary

Problems solved by technology

[0004] Despite great progress in graph learning in multi-view clustering, its performance in multi-stage processes still shows significant limitations: in such a framework, the optimization of the objective function for each stage is independent , thus reducing the final clustering performance

For a specific objective, the graphs for each view are independently learned in a linear space; to fuse graphs from multiple views, additional parameters are usually required, and the choice of parameters affects the clustering performance, making it difficult to simultaneously learn all Valid for certain tasks

Images