An Improved Method to Reduce Over-Segmentation of Watershed Transformation and its Application in the Contour Extraction of Brain Image
Hui Zhu, Bofeng Zhang, Anping Song, Wu Zhang · 2009
Watershed transformation is a common technique for image segmentation. However, its use for medical image segmentation has been limited particularly due to over-segmentation. In response to the characteristics of medical image, especially the contour extraction from the MRI (magnetic resonance imaging) brain image, this paper proposes an improved method in order to overcome the drawbacks. Firstly, multi-scale alternating sequential filtering by reconstruction is introduced to eliminate the noise and simplify the input images, and the loss of boundary information can be avoided. Secondly, two methods of h-minima and minima imposition are imposed on the gradient image to mark the minima regions, so all its local minima are suppressed. Finally, the watershed algorithm is applied to the marked gradient images to get the contour of brain. Experimental results show that the improved method can be applied to contour extraction of MRI brain image with good result, and the mean reduction of local minima in the over-segmented image of regions is 68.26% compared to the watershed transformation based on mark extraction.