Change Smoothness-Based Signal Decomposition Method for Multimodal Change Detection

Abstract

Change detection using multimodal remote sensing images is a very important and challenging topic. Due to the different imaging conditions, multimodal images cannot be directly compared to obtain changes. To address this challenge, in this letter we propose a change smoothness based signal decomposition (CSSD) model to decompose the post-event image into a regression image of pre-event image and a changed image. By establishing the pairwise relationship between the superpixels within the image, we construct a graph for the changed image and use the edge weights to measure the state consistency of connected vertexes, i.e., the probability of being both changed or unchanged. We show that the changed image is smooth on the constructed graph. In contrast to the previous methods that only use the prior change sparsity, we also exploit the change smoothness in the signal decomposition model, which makes the CSSD more robust and accurate, and outputs a better changed image, thus improving the change detection performance. Experimental results and comparisons with seven state-of-the-art methods on three datasets demonstrate the effectiveness of the proposed method.