Combining different visual vocabularies with different sizes for image categorization

In this paper, the advantages of ensemble methods are applied to image categorization. A novel method is introduced for image categorization by combining various visual vocabularies with different sizes in the popular vocabulary approach. The vocabulary approach describes an image as a bag of discrete visual codewords, where the frequency distributions of these words are used for image categorization. Based on vocabularies of various sizes, a classifier ensemble is learned, which can jointly exploit different information with various granularities. High classification accuracies of the proposed algorithm are demonstrated on four different datasets.