Abstract
A contrast-sensitive Potts model (CSP) custom-designed for change detection is presented using remotely sensed images. In traditional Potts model, a constant penalty coefficient is used, which results in ignorance of significant details and excessively homogenous patches during change detection using the difference image generated from multitemporal images. In the proposed CSP, the difference image is divided into unchanged, uncertainty and changed regions. Then different linear functions are introduced instead of the constant penalty coefficient for different regions. Two experiments were carried on optical satellite images, and the results indicate that the proposed CSP produces more accurate change maps than some state-of-the-art methods.