Dense sampling low-level statistics of local features

Abstract

Recently, generic image recognition techniques are widely studied for automatic image indexing. However, much of these works are computationally too heavy for practical large setup. Thus, it is very important to properly balance the trade-off between performance and computational cost for realizing scalability. In recent years, methods based on the bag-of-keypoints technique have been quite successful and are widely used. However, preprocessing cost for building visual words becomes immense in large scale datasets. On the other hand, methods based on global image features have been used for a long time. Because global image features can be extracted rapidly, it is relatively easy to use them with very large datasets. However, the performance of global feature methods is usually poor compared to bag-of-keypoints. In this paper, we propose a very simple but powerful scheme of boosting the performance of global image features, by densely sampling low-level statistics (mean and correlation) of local features. Also, we use a highly scalable learning and classification method which is substantially lighter than SVM. Our method achieved the performance comparable to state-of-the-art methods in spite of its remarkable simplicity.