Change Detection in Urban Area from High Resolution Satellite Images Using Local Features

The automatic change detection in urban environment using multitemporal high resolution satellite images is one of the fundamental processes in photogrammetry and remote sensing. Conventional approaches to automated building change detection relied to multitemporal images comparison based on multispectral classification methods. These approaches require accurate alignment between the two images to detect changes. Accurate registration of high resolution satellite images is a difficult process due to local distortion in these images due to relife displatement. This paper introduces a multi temporal image processing approach towards an efficient and automated detection of urban changes. The proposed method is based on local features obtained from sequential images including corners, blobs and regions. In the first step, local features are extracted in each of the images using three well-known feature extractor operator incuding phase congruency, UR-SIFT method and MSER algorithm. Then, feature matching process in a gridding structure is performed between two feature sets using SIFT descriptor. This process followd by a new mismatch elimination method based on distance ratio. In the next step, a similarity image based on a new measure is estimated between multitemporal images. In final, using an automatic threshold changed regions are determined. The method has been evaluated by using QuickBird and World Wiew image for the area of the city of Sun francisco, USA. Experimental results prove the capabilities of the proposed change detection algorithm with appropriate accuracy in multitemporal high resolution satellite images.