Image segmentation based on NSCT and watershed

A new method for image segmentation is presented. It combines the nonsubsampled contourlet transform (NSCT) with the marker-controlled watershed algorithm to control over-segmentation and therefore achieving better segmentation. The NSCT is employed to denoise the image and enhance the edges in multiple resolutions, and the image gradient is estimated at each scale. The marker-controlled watershed algorithm is then applied to the obtained gradient images. Finally, the labeled bandpass images and the corresponding high bandpass images are constructed by the inverse NSCT to obtain the final segmented image. Results of the experimental study indicate that the proposed method is a practicable approach to control over-segmentation and can segment objective accurately.