An improved image classification based on K-means clustering and BoW model

Abstract

Image classification constitutes an important issue in large-scale image data process systems based on cluster. In this context, a significant number of relying BoW models and SVM methods have been proposed for image fusion systems. Some works classified these methods into Generative Mode and Discriminative Mode. Very few works deal with a classifier based on the fusion of these modes when building an image classification system. In this paper, we propose a revised algorithm based on weighted visual dictionary of K-means cluster. First, it uses SIFT and Laplace spectrum features to cluster object respectively to get local characteristics of low dimension images (sub-visual dictionary); then clusters low-dimension characteristics to get the super visual dictionaries of two features; finally, we get the visual dictionary although most of these features have been proposed for a balance role through weighting of the parent visual dictionaries. Experimental result shows that the algorithm and this model are efficient in descript image information and can provide image classification performance. It is widely used in unmanned-navigation and the machine-vision and other fields.