Energy minimization model for image segmentation via graph cut optimization
Xiantao Liu · Jisuanji yingyong yanjiu · 2012
Aiming at the drawbacks of active contour models which used gradient descent and result in local minimum easily,this paper proposed a discrete energy minimization model for image segmentation.It designed the new model based on Chan-Vese model and optimized the energy function via graph cut method.It could find a global minimum rather than a local one.To construct the new model,first step was to map the image for a graph,and then changed the level set energy function into a discrete form which should be proved graph-representable.Using the model traversed each node and its neighborhood,it computed the weights of the edges with these nodes,and got the new labels of pixels and updated the initial contour,until the energy remained constant and the contour reached the boundary of object.The major advantages of this model included the existence of global minimum and its insensitivity to initialization.Numerical implementations show that the model improves the accuracy and speeds for image segmentation.