To improve the segmentation accuracy of mycobacterium tuberculosis (MTB) objects, a segmentation algorithm for optical microscopic MTB images in region level was proposed. Top-bottom hat transform was used to enhance the contrast of the color images, and the image gradients were computed by comprising local features and global information of the images. Watershed algorithm was employed to implement the initial segmentation. Segmented regions were then merged by using the maximum similarity criterion in adjacent regions in order to obtain integrated object regions. The method of multi-threshold segmentation in terms of the color characteristics of MTB object regions was adopted to filter the impurities and to realize the segmentation of MTB objects. Experimental results indicate that the proposed algorithm can segment MTB objects which have low contrast and saturation and can obtain well-segmented results for images in different dyeing backgrounds.