This document summarizes a proposed method for interactive change detection in multispectral remote sensing images. The method uses user-provided markers for change and no-change classes to train a support vector machine classifier, generating a spectral change map. Two approaches are described to incorporate spatial context - a Markov random field method that is region-driven, and a level-set method that exploits both regions and contours. Experiments on real remote sensing images showing different types of changes demonstrate that the interactive methods can accurately generate change detection maps with simple minimal user interaction.