Segmentation of boron carbide microscopic images which present twins

Boron carbide microscopie images may present grains crossed by twins (or macles) viewed as straight lines in the inner part of each grain. The twins preclude the segmentation because they are very similar to the grain borders. To eliminate the twins a preprocessing step, which relies on mathematical morphology tools, is applied to the image. The main operation used is a directional opening, using structuring elements with several sizes and directions. After this step the image is ready to be segmented. A modified watershed algorithm, CBMA, is then used to segment the image. This algorithm employs two geometric attributes: an area and a depth criteria to merge the non-significant catchment basins.