Multivariate statistical modeling for multi-temporal SAR change detection using wavelet transforms

In this paper, we propose a new method for automatic change detection in multi-temporal SAR images based on statistical wavelet subband modeling. The image is decomposed into multiple scales using wavelet transform and the probability density function of the sliding window coefficients of each subband is assumed to be multivariate Gaussian distribution. Kullback-Leibler similarity measures are computed between two corresponding subbands of the same scale and used to generate the change map. The multivariate statistical model is considered here to better model the spatial information given by texture than that given by a univariate statistical model. The proposed method is compared to the classical method based on univariate Gaussian distribution. Test on real data show that our approach outperforms the conventional approach.