Detecting change of remote sensing images is very important for some applications such as tracking of moving objects and motion estimation. Traditional work on change detection has largely been based on segmentation approaches of a single feature. It excessively depends on the threshold of the single feature to determine whether the change of spectral information is caused by the change of object. The results of traditional change detection approaches can easily be affected by noise, blur, contrast level and brightness level. To overcome the deficiency, we improve the Itti visual saliency model and propose an effective and robust approach based on saliency map to detect real changed regions between two remote sensing images of a given scene acquired at different times. The results of the experiments indicate that our approach is very robust to noise, contrast level and brightness level.