A classic approach to obtain High Dynamic Range Images (HDRI) consists in combining multiple images of the same scene with varying exposures. However, if the scene is not static during the time of capture, moving objects will appear blurry and ghosted, i.e. in multiple locations. Detecting and removing ghosting artifacts is an important issue for automatic generation of HDRI of dynamic scenes. This paper first describes a new method for detecting regions where ghosting occur based on an order relation between pixel values in consecutive images. Secondly, we propose to use a robust statistics tool to remove the detected ghosts. Results with different scenes and comparison with two other methods show the effectiveness of the approach.