Research On Image Clustering Algorithm Based On Bi-graph Regularization Concept Factorization

In an era of rapid technological development,there are various ways and means for humans to obtain data,and at the same time,more and more problems are caused.The shortage of storage space and the consumption of running time caused by massive data have brought serious challenges to scientific research.How to propose a dimensionality reduction algorithm to keep the original high-dimensional space characteristics of the data in the reduced dimensionality space is the focus and difficulty of dimensionality reduction algorithm research.Non-negative matrix factorization(NMF),which is popular in recent years,is a highly explanatory dimensionality reduction technique and is widely used in the field of image clustering.Based on NMF,we propose two NMF models by depicting the internal geometric structure of data distribution as follows:The proposed Three-Graph regularization Concept Factorization(GTCF)algorithm takes Concept Factorization(CF)as the main body,and combines regularization terms of feature manifold,data manifold regularization,and self-similar learning regularization.The GTCF algorithm considers the characteristics of basis vectors and sparse vectors at the same time,and better describes the intrinsic geometric structure of the data space.CF as the main body enhances the sparsity of the algorithm.The six algorithms are compared on three data sets to show that GTCF has a good effect on improving the clustering performance.The convergence of GTCF is demonstrated through theory and experiments.The proposed Dual Graph Global and Local Concept Factorization(DGLCF)describes the complex internal global and local manifold structure of the data space and helps to characterize the complex manifold space.DGLCF introduces the global and local structure of the data manifold and the geometric structure of the feature manifold into the concept Factorization(CF).The global manifold structure makes the model more discriminative,while the two local regularization terms preserve the inherent geometric structure of the data and features at the same time.Finally,the convergence and iterative update rules of DGLCF are analyzed by The comparison with six clustering algorithms in a real dataset shows the advantages of clustering performance.