A segmentation algorithm for Mycobacterium Tuberculosis images based on automatic-marker watershed transform

In order to avoid the over-segmentation problem caused by original watershed transform and improve the segmentation precision of Mycobacterium Tuberculosis (MTB) images, a novel segmentation algorithm is proposed based on automatic-marker watershed transform. The automatic marker is accomplished by Gaussian weighted adaptive threshold segmentation and local minimum points search within gradient image. After the stage of initial segmentation based on watershed, regions are merged with the criterion of max similarity of adjacent regions in order to obtain integrated objects. Finally, the multi-thresholds segmentation for eliminating contaminations is employed to improve the robustness of the algorithm. Experimental results demonstrated that the proposed algorithm can achieve superior segmentation accuracy in the images with different background colors.