Image classification with Bag-of-Words model based on improved SIFT algorithm

The common method of image classification based on traditional SIFT local feature description makes the description of the global information not comprehensive and has complicated calculation because of the construction of scale extreme space. In addition, the feature space is high dimensional and sparse which will result in low classification accuracy, data redundancy and time-consuming process. The paper adopts a new image classification method with Bag-of-Words model based on improved SIFT algorithm. Each image is divided into a lot of uniform grid patches and the single scale SIFT feature descriptor with 128 dimensional is extracted in each patch. Then combine the PCA theory to reduce the dimensions of SIFT feature vector from 128 d to 20 d. Next, the BOW model of the image will be obtained by visual vocabulary. Finally establish the support vector machine (SVM) classifier based on radial basis function (RBF) and histogram intersection kernel (HIK) function respectively with the data above for training and testing. The optimal scheme is concluded through comparison of experimental results. The experimental results show that, the method presented in this paper shows higher classification accuracy.