Digital change detection with the aid of multiresolution wavelet analysis

This Letter presents the preliminary findings of a new approach to deal with misregistration effects on change detection results. A multiresolution analysis with wavelet transforms applied to image differencing results enabled the extraction of changed sites according to size classes. Changes of interest were pinpointed successfully without the necessity of accurate spatial registration or radiometric rectification while differences not related to land cover changes were bypassed. The method's applicability is demonstrated with a multitemporal data set of Landsat MSS and TM images for the detection of deforestation and new areas of rock exploitation in south-eastern Brazil.