Generating Discriminative Visual Vocabulary Based on Fusion of Features

Constructing a discriminative visual vocabulary is important in image categorization. Most visual vocabularies are built on a clustering method such as K-Means to cluster feature vectors of image patches. And the clustering methods ignore the different impacts of features to build a more discriminative visual vocabulary. In this paper a discriminative visual vocabulary generating method based on fusion of multiple features is proposed. The Dempster-Shafer (D-S) evidence theory is applied to explore the visual similarity among different features so that features have different effects on clustering to achieve a more discriminative visual vocabulary. The experimental results showed that our method achieved better classification accuracies of image categorization on both the Support Vector Machine (SVM) classifier and the Naive Bayes (NB) classifier. It demonstrated that our visual vocabulary is more discriminative than the vocabulary built on K-Means clustering method.