The main objective of this letter is to improve the accuracy of unsupervised change detection method and minimize registration errors among multi-temporal images in the change detection process. To this end, iteratively regularized multivariate alteration detection (IR-MAD) is applied to synthetically fused images. First, four synthetically fused hyperspectral images are generated using the block-based fusion method. Then, the IR-MAD is applied to three pairs of the fused images using integrated IR-MAD variates, to decrease the falsely detected changes. To focus on the mis-registration effects, we apply the method to both a correctly registered data-set and a data-set with deliberately misaligned images. In this experiment using multi-temporal Hyperion images, the changed areas are more efficiently detected by our method than by the original IR-MAD algorithm.