Optimizing Decision for Scene Classification Based on Latent Semantic Analysis

In this paper,we propose a scene classification method based on latent semantic analysis using some priori information to overcome the limitation of semantic modeling and solution of parameters in conventional pLSA models.First,the optimization decision model is built with the summation of two prior constraint conditions simultaneously on a traditional parameter matrix: low rank for the database and sparsity for the individual images.Second,the decision model is solved efficiently by using the method of inexact augmented Lagrange multiplier.Finally,a series of experiments and comparative analysis are carried out with the proposed scene classification approach in different scale scene datasets.Experimental results show that our method can provide a compact semantic representation in a low dimension space and deal with the problems of local optimum with good convergence.