Semantic image segmentation integrating gradient reconstruction and non-linear processing

Liu Ji-lin,U Biometrics
2008-01-01
Abstract:Semantic image segmentation provides significant foundation for image analysis and understanding. Oriented to semantic segmentation, this paper proposes an image segmentation method integrating grad reconstruction and non-linear processing, aiming at resolving the problems of sensitivity to noise and over-segmentation existing in traditional watershed algorithm. At first, a method based on double morphological opening and closing reconstruction is developed to attain morphological gradient, which passes through a certain threshold transform to restrain light and dark details caused by noise. Subsequently, a non-linear classification based on certain scale is employed to reduce the influence of close texture insides objects and background. Finally, an improved watershed labeling algorithm is proposed and performed. Experimental results illustrate that the proposed method is superior to traditional method and that used in Matlab toolboxes.
What problem does this paper attempt to address?