Scene Classification using Generalized Local Correlation
Hideki Nakayama, Tatsuya Harada, Yasuo Kuniyoshi · 2009
Feature extraction is an important issue for generic image recognition. In recent years, methods based on the bag-of-keypoints technique have been quite successful and are widely used. However, this technique requires the quantization of local patches to build visual words as a preprocessing step, the computational cost of which is enormous. 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 in practical large-scale systems. However, the performance of global feature methods is usually poor compared to bag-of-keypoints. Therefore, it is essential to develop a more powerful scheme of global feature extraction for achieving practical applications of generic image recognition. In this paper, we show that we can boost the performance of global image features by considering the correlations of local features in addition to the mean. We experimentally verify the effectiveness of our method using standard scene classification benchmark datasets. 1