A wavelet-based change-detection technique for multitemporal SAR images

This paper presents a novel wavelet-based multiscale technique for unsupervised change detection in multitemporal synthetic aperture radar (SAR) images. The proposed approach is based on the analysis of a set of scale-dependent images characterized by a different trade-off between speckle reduction and preservation of geometrical details. The different scales are obtained by means of a multiresolution decomposition of the log- ratio image (obtained by a comparison of a pair of co-registered images acquired at different times on the same area). The final change-detection map is derived according to an adaptive scale- driven fusion algorithm, which properly exploits information at different resolution levels. According to an automatic local analysis of the statistic of the data, for each pixel only a sub-set of reliable scales is selected and exploited in the decision process thus producing an accurate and reliable change-detection map in both homogeneous and border areas. Experimental results confirm the effectiveness of the proposed technique. In order to address the above limitations of the standard methods, in this paper we present a novel approach to change detection in multitemporal SAR images. The proposed approach exploits a wavelet-based multiscale decomposition of the log-ratio image (obtained by a comparison of the original multitemporal data) aimed at achieving different scales (levels) of representation of the changed areas. Each scale is characterized by a different trade-off between speckle reduction and preservation of geometrical details. Then scale- dependent log-ratio images are analyzed to obtain the final change-detection result according to an adaptive scale-driven fusion algorithm. The fusion step aims at properly exploiting the different behaviors at different scales for producing an accurate and reliable change-detection map. In greater detail, a set of reliable resolution levels is defined according to an adaptive comparison between the pixel local statistics and global statistics independently performed at each scale. A scale-driven fusion strategy is applied at decision or feature level to compute the final change-detection map. The basic idea is to use high-resolution levels only in the analysis of the expected edge (or details) pixels and to consider also low- resolution levels in the processing of pixels in homogeneous areas. Thus, the proposed method exhibits both a high sensitivity to geometrical details (e.g., border of changed area are well preserved) and a high robustness to speckle noise in homogeneous areas.