The high spatial resolution remotely sensed imagery has abundant detailed information of earth surface, and the multi-temporal change detection for the high resolution remotely sensed imagery can realize the variations of geographical unit. In terms of the high spatial resolution remotely sensed imagery, the traditional remote sensing change detection algorithms have obvious defects. In this paper, learning from the object-based image analysis idea, we proposed a semi-automatic threshold selection algorithm named OB-HMAD (object-based-hybrid-MAD), on the basis of object-based image analysis and multivariate alternative detection algorithm (MAD). which used the spectral features of remotely sensed imagery into the field of object-based change detection. Additionally, OB-HMAD algorithm has been compared with other the threshold segmentation algorithms by the change detection experiment. Firstly, we obtained the image object by the multi-solution segmentation algorithm. Secondly, we got the object-based difference image object using MAD and minimum noise fraction rotation (MNF) for improving the SNR of the image object. Then, the change objects or area are classified using histogram curvature analysis (HCA) method for the semi-automatic threshold selection, which determined the threshold by calculated the maximum value of curvature of the histogram, so the HCA algorithm has better automation than other threshold segmentation algorithms. Finally, the change detection results are validated using confusion matrix with the field sample data. Worldview-2 imagery of 2012 and 2013 in case study of Beijing were used to validate the proposed OB-HMAD algorithm. The experiment results indicated that OB-HMAD algorithm which integrated the multi-channel spectral information could be effectively used in multi-temporal high resolution remotely sensed imagery change detection, and it has basically solved the “salt and pepper” problem which always exists in the pixel-based change detection, and has mitigated the impact of building shadows and geometric registration error, and has improved the overall accuracy and kappa coefficient than other change detection algorithm, but it has more undetected error. By compared with the SNR of image object, we know that the MNF transformation could effectively improve to concentrate the change information.