Fusion of multiple VLAD vectors based on different features for image retrieval

Traditional VLAD method only uses the SITF feature. Since the SITF feature represents the local gradient information, thus VLAD representation based on SITF feature of image has low discriminative power. To address the problem, we present a simple and effective method that fuse the VLAD vectors based on local gradient and color information. Also, in order to improve the retrieval accuracy and reduce running time, we use whitening operator for VLAD vectors. Our proposed method is evaluated on two benchmark datasets, i.e., Holidays and Ukbench. Experiments show that our proposed method achieves good performance.