Curvelet-based Change Detection for man-made Objects from SAR Images

Abstract

In this article we present a technique for fast and robust change detection based on the curvelet transform. The curvelet transform is a two dimensional further development of the well-known wavelet transform that reconstructs the original image by ridge-like features, called ridgelets, in different scales, directions and positions. Geocoded SAR amplitude images from TerraSAR-X are compared by differentiating the coefficients of both images in the curvelet coefficient domain. Before the difference image is transformed back to the spatial domain, the influence of the single ridgelets on the resulting image can be manipulated to suppress noise and to intensify structures. Two examples were chosen to show the potential of this approach: a construction site in Germany and an open cast mining area in Chile. Our prototype version is able to compare time series without any interaction of an operator so that the implemented algorithms can easily be embedded into an automatic processing chain.