This paper proposes a novel automatic change detection approach for single band multi-temporal remote sensing images (MTRSI). First, the difference image is constructed by combining the spatial neighborhood information with the improved multiplying transform fusion (MTF) technique, which can well weaken noises and eliminate the effects caused by the registration error of multi-temporal images. In the segmentation processing of the difference image, the distributions of changed and unchanged classes are fitted by Rayleigh-Gauss models (RGM) and the probability densities of changed and unchanged pixels are estimated. Then the optimal change detection threshold is calculated automatically and effectively by the improved Kittler-Illingworth (KI) threshold selection algorithm. Finally, the changed regions are extracted. The experimental results obtained on the simulated MTRSI and the real MTRSI confirmed the effectiveness of the proposed approach. In particular, the results in terms of overall error and overall detected accuracy proved that the proposed generation approach of the difference image could have better performance than the MTF technique. In addition, as expected, the RGM was proved to be more suitable than the Gauss models (GM) and the Generalized-Gauss models (GGM) to fit the distributions of changed and unchanged classes. And the change detection experiments also confirmed that the proposed automatic threshold selection method based on RGM fitting technique could achieve the very similar performance to the optimal results exhibited by the supervised manual trial and error procedure (MTEP). This paper proposes a novel automatic change detection approach for single band multi-temporal remote sensing images (MTRSI). First, the difference image is constructed by combining the spatial neighborhood information with the improved multiplying transform fusion (MTF) technique, which can well weaken noises and eliminate the effects caused by the registration error of multi-temporal images. In the segmentation processing of the difference image, the distributions of changed and unchanged classes are fitted by Rayleigh-Gauss models (RGM) and the probability densities of changed and unchanged Then the optimal change detection threshold is calculated automatically and effectively by the improved Kittler-Illingworth (KI) threshold selection algorithm. Finally, the changed regions are extracted. The experimental results obtained on the simulated MTRSI and the real MTRSI confirmed the effectiveness of the proposed approach. In particular, the results in terms of over all error and overall detected accuracy demonstrated that the proposed generation approach of the difference image could have better performance than the MTF technique. In addition, as expected, the RGM was proved to be more suitable than the Gauss models (GM) and the Generalized- And the change detection experiments yet also that that proposed novel automatic threshold selection method based on RGM fitting technique could achieve the very similar performance to the optimal results attracted by the supervised manual trial and error procedure (MTEP).