Efficient Computation of Morphological Greyscale Reconstruction

Morphological reconstruction is an important image operator from mathematical morphology. It is very often used for filtering, segmentation, and feature extraction. However, its computation can be very time-consuming for some input data. In this paper we review several efficient algorithms to compute the reconstruction, and compare their performance on real 3D images of large sizes. Furthermore, we propose a GPU implementation which performs up to 15× faster than the CPU methods. To our best knowledge, this is the first GPU implementation of the morphological reconstruction, described in literature. Digital Object Identifier 10.4230/OASIcs.MEMICS.2010.54