Viscous morphological gradient modification for watershed segmentation

Watershed transform is an efficient way to segment image; however, it often results in over-segmentation. Usually, smoothing pre-filtering is adopted to avoid over-segmentation, but the positions of the segmented regional contours may be incorrect. For the purpose of segmenting image without over-segmentation and without position errors of regional contours, a method for watershed segmentation is proposed based on viscous morphological gradient modification. First, original image is transformed by morphological gradient transform, and then viscous morphological operators with structure elements of different sizes are used to modify the relief surface of gradient image; after modification, most irregular local minimums caused by details and noise are removed, while positions of regional contours are not or slightly changed due to the less modification of the high gradient levels. Finally, standard watershed transform is employed to implement segmentation on the basis of modified gradient image. Experiments show that this method not only effectively avoids the over-segmentation of watershed, but also preserves the positions of regional contours.