High spatial resolution (HSR) image scene classification is aimed at bridging the semantic gap between low-level features and high-level semantic concepts, which is a challenging task due to the complex distribution of ground objects in HSR images. Scene classification based on the bag-of-visual-words (BOVW) model is one of the most successful ways to acquire the high-level semantic concepts. However, the BOVW model assigns local low-level features to their closest visual words in the “visual vocabulary” (the codebook obtained by k
-means clustering), which discards too many useful details of the low-level features in HSR images. In this paper, a feature coding method under the Fisher kernel (FK) coding framework is introduced to extend the BOVW model by characterizing the low-level features with a gradient vector instead of the count statistics in the BOVW model, which results in a significant decrease in the codebook size and an acceleration of the codebook learning process. By considering the differences in the distributions of the ground objects in different regions of the images, local FK (LFK) is proposed for the HSR image scene classification method. The experimental results show that the proposed scene classification methods under the FK coding framework can greatly reduce the computational cost, and can obtain a better scene classification accuracy than the methods based on the traditional BOVW model.
This is an open access article distributed under the Creative Commons Attribution License
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited