To deal with the problem that changed areas are overestimated in image change detection, a context-sensitive similarity based supervised image change detection method is proposed and applied in bi-temporal high resolution and multispectral satellite image change detection. It exploits both context-sensitive magnitude and direction of change in the vicinity of each pixel by means of local intercept and slope, and uses support vector machine (SVM) with local intercept and slope for image change detection. The results are from experiments for change detection of high resolution bi-temporal multispectral earthquake images including building damage. It is obvious that the false alarms are mostly reduced, which means it is effective for solving changed area overestimation problems.