Robust change detection in dense urban areas via SVM classifier

This paper introduces a novel unified framework for change detection in remote sensing images, which compute one local dHOG feature from two images and make classification based on SVM classifier. Compared to the traditional methods, this approach takes advantage of the robustness of the dHOG feature. The inaccuracy and ambiguity with the definition of change can be eliminated by SVM classifier by training with an expert labeled dataset. In order to tackle the projective deformation problem which usually produce substantive false alarms, a novel matching algorithm is introduced by solving a discrete optimization problem. Experiments demonstrate the advantages and effectiveness of the proposed method.