Resolution selective change detection in satellite images

In this paper, we propose a novel method for unsupervised change detection in satellite images. A feature vector for each pixel is extracted using the multiresolution representation of the difference image which is computed from the multitemporal satellite images of the same scene acquired at different time instances. A metric for automatically estimating the number of resolution levels used in multiresolution analysis is proposed. The dimensionality of each feature vector is reduced using principal component analysis (PCA). The feature vectors are then classified into “changed” and “unchanged” classes using k-means clustering with k = 2 to achieve a change detection map. Results are shown on real data and comparisons with the state-of-the-art techniques on advanced synthetic aperture radar (ASAR) images are provided.