Multi-focus image fusion based on fractional order differentiation and closed image matting

This paper presents a novel approach to addressing the fusion of multi-focus images in either registered or mis-registered cases. The conventional approaches often produce blurred edges of objects in the fused images due to inaccurate decision maps. On the other hand, these decision maps are sensitive to mis-registration that causes artifact in the fused images. Therefore, we propose a robust multi-focus image fusion approach with clear object edges for the registered or mis-registered source images. In this approach, a fractional order differential mask is creatively adopted to pre-process the source images, ensuring the initial decision maps both with the boundaries and fine structures of the objects and with the internal holes closed. Then, the closed matting technique, in lieu of the robust matting, is adopted to refine the initial decision maps. This significantly reduces the interaction information from the users, but still preserves the complete boundaries of the objects. Finally, the global threshold processing is skillfully adopted to form the decision maps. This not only yields the final decision maps with smooth boundaries, but also guarantees the rich gradient information from the mis-registered source images. The experimental results show that the designed algorithm provides better visual perception and higher objective evaluation than some existing representative algorithms.