Reducing the effects of misregistration on pixel-level change detection

A model that compensates for misregistration effects on change detection results shows promise for reducing artefacts and enhancing land change features at or near the pixel scale and for reducing noise caused by misregistered muti-temporal images. Sparse estimates of misregistration across the scene are combinedwithcalculations of spatial brightness gradients toadjust the magnitude of multi-temporal image differences. The model is tested on a multi-temporal Landsat Thematic Mapper image data set for a rapidly urbanizing landscape in southern California.