Image Segmentation Based On Watershed Algorithm

Image segmentation is one kind of highly specialized processing techniques that extract important information from the image.The effect of image segmentation will affect the cost of next steps.Currently image segmentation has been a very big breakthrough,and many excellent methods in accuracy and the target object boundary segmentation better fit.While oversegmentation and under-segmentation are still serious problems in current society.Watershed algorithm was a geodesic topological image segmentation technologies.Due to the influences of the nose and the gradient images' detail textures,watershed segmentation algorithm appeared over-segmentation phenomenon,affecting the result of segmentation.In order to improve the segmentation effect and take into account regional consolidation process' complex and large scale of computing,two algorithms were proposed in this paper,which preprocessing the image before the watershed algorithm.These algorithms were watershed algorithm based on multi-scale morphological markers and watershed algorithm based on texture feature and kernel clustering binding.The watershed algorithm based on multiscale morphological markers were proposed in the paper.Firstly,the gradient image was marked by the transformation of H-minima;then the gradient image was reconstructed by the multi-scale structure elements;each reconstructed gradient and gradient image were subtracted respectively.The low gray areas which were smaller than the structure element were filled.The difference was marked as an image and the union mark images were used as a new marker.Finally,the area thresholds were used to remove the spots of the new marker image.The watershed algorithm was used to deal with the gradient image which was modified by the forced minimum technique.By comparing the simulation results,this algorithm was proved to be a good method.The watershed algorithm based on texture feature and kernel clustering was proposed in the paper.Firstly,the RGB space was converted into L*A*B color space;Secondly,clustered similar texture,the region blocks with similar texture features were combined and classified;then the color point of view was clustered.Fuzzy kernel clustering was used to merge similar pixels in the L*A*B channel;finally,the watershed algorithm was used to deal with the gradient image;the simulation results confirm that this algorithm was proved to be a good method.