Feature selection based codebooks construction for scene categorization

Wenjie Xie,De Xu,Songhe Feng,Yingjun Tang

DOI: https://doi.org/10.1109/ICOSP.2010.5655717

IF: 4.729

2010-01-01

Signal Processing

Abstract:In scene categorization, one single histogram based on the sole universal codebook is used to characterize an image in most state-of-the-art scene categorization methods, which is lack of enough discriminative ability to separate the images among different categories and results in low classification accuracy. In order to solve the problem, in this paper, we propose a novel scene categorization approach that constructs class-specific codebooks based on feature selection method. In our proposed approach, feature selection method is adopted to measure the visual word's contribution and construct class-specific codebook for each category. Then, an image is characterized by a set of combined histograms (one histogram per class) which are generated by concentrating the traditional histogram based on universal codebook and the class-specific histogram grounded on class-specific codebook with an adaptive weighting coefficient. The improved combined-histogram provides useful information or cue to overcome the similarity of inter-class images. Experimental results on Lazebnik 15 dataset show that our proposed scene categorization method significantly outperforms the state-of-the-art approaches.