In this paper, a novel bag of visual words algorithm for object detection in satellite images is presented. Two main extensions to classical bag of visual words algorithm are proposed: Exploiting scale information and defining important visual words. Although bag of visual words model is widely used, there isn't enough study about using scale information and weighting visual words due to an importance measure. A new descriptor is presented by adding scale information to SIFT descriptor and a novel visual word weighting algorithm is proposed considering that more occurrence in the object and less in the background is an importance measure.
展开▼