An Image Classification Method Based on PLSA and Visual Phrases

Abstract

In the traditional image classification methods based on the Spatial Pyramid Model (SPM), SPM increased the computational complexity of image classification in the process of obtaining the spatial distribution information of the local features, and the sub-region of SPM is defined in advance, so the image classification methods based on SPM have some limitations. In view of the problems above, this paper proposed an image classification method based on probabilistic latent semantic analysis (PLSA) and visual phrases. Firstly, the method obtained the spatial distribution information of the local features by establishing visual phrases. Then, a new semantic visual dictionary was constructed based on the visual phrases. Finally, the PLSA was used to model all images. Experiments on two common image databases show that the image classification performances of the proposed method is significantly improved to some extend compared with another two traditional methods.