SCENE CLASSIFICATION USING CASCADED PROBABILISTIC LATENT SEMANTIC ANALYSIS

In this paper we propose a novel approach of image representation for weakly supervised scene classification that mainly combine two popular methods in the literature: Bag-of-Words (BoW) modeling and probabilistic Latent Semantic Analysis (pLSA) modeling. The new image representation scheme called Cascaded pLSA performs pLSA in a hierarchical sense after the BoW representation based on SIFT features is extracted. We associate location information with the conventional BoW/pLSA algorithm by subdividing each image into sub-regions iteratively at different resolution levels and implementing a pLSA model for each sub-region individually. Finally, an image is represented by concatenated topic distributions of each sub-region. The performance of our method is compared with the most successful methods in the literature using the same dataset. In the experiments, it has been seen that the proposed method outperforms the others in that particular dataset.