SIFT features of fusion region information entropy in Bag-of-Words

In the object recognition based on the bag of visual words, we focus on several main feature extraction algorithms. The Scale Invariant Feature Transform (SIFT) that based on local features with good significance and robustness has became a popular feature extraction method. But we take into consideration that the SIFT will generate a large number of high-dimensional feature vectors, which will increase the computational cost. In this paper, a novel feature extraction algorithm by means of SIFT to fuse region information entropy (SIFT-entropy) is proposed. The algorithm can improve the quality of the feature extracted from the image by cluster similar key points to removing the noisy key points. So a more discriminative high quality “visual word” codebook could be generated. We made a comprehensive comparison between the proposed method and the original SIFT method on Caltech-101 database. The experimental results show that this improvement can reduce the dimension of the feature space, and has higher classification accuracy.