Unsupervised Modeling and Recognition of Object Categories with Combination of Visual Contents and Geometric Similarity Links

Abstract

This paper proposes a probabilistic approach for unsupervised modeling and recognition of object categories which combines two types of complementary visual evidence, visual contents and inter-connected links between the images. By doing so, our approach not only increases modeling and recognition performance but also provides possible solutions to several problems including modeling of geometric information, computational complexity, and the inherent ambiguity of visual words. Our approach can be incorporated in any generative models, but here we consider two popular models, pLSA and LDA. Experimental results show that the topic models updated by adding link analysis terms significantly improve the standard pLSA and LDA models. Furthermore, we presented competitive performances on unsupervised modeling, ranking of training images, classification of unseen images, and localization tasks with MSRC and PASCAL2005 datasets.