This chapter discusses the context of unsupervised change detection using bitemporal synthetic aperture radar (SAR) images. It aims that to perform texture tracking based on the characteristic points extracted from the images and modeled by a graph structure. Many methods have been proposed to tackle the problem of image change detection. In the scope of unsupervised SAR image change detection, one of the most classical approaches is the mean-ratio detector which is based on the ratio of local intensity means of pixel patches. The chapter presents our motivation and approach to exploit local extrema keypoints to represent textures in remote sensing images, and the adaptation to SAR data. The non-local mean model involves a filtering process to reduce speckle noise during the change detection algorithm. The chapter also presents a keypoint graph-based texture tracking method to tackle unsupervised change detection and glacier displacement measurement using bitemporal SAR images.