Image Segmentation Methods Based On Active Contour Combined With Superpixel Force And Generative Method

Image segmentation is a basic problem in computer vision and image processing.Image segmentation refers to divide image into several disjoint areas according to the similarity criterion,including similar gray scale,color,texture and other similar characterist.How to improve the segmen-tation precision,reduce the time of segmentation and avoid ambiguity has been the focus of the research field.For this reason,scholars have pro-posed a series of segmentation methods.The Gaussian mixture model and the active contour model are two typical segmentation technologies,ex-hibiting complementariness in both the involved objective functions and segmentation results.In this paper,we fistly propose a new segmentation method based on regional statistic information and the geometric infor-mation of the active contour.Our segmentation method enables the active contour model contains information of probability items in the evolution process,while probabilistic items have geometric information of active contour when updating parameters,so as to achieve good segmentation ef-fect.And then,we introduce labelled samples from moderate user interac-tion,and use the semi-supervised generative method to improve the model,which can significantly avoid ambiguity of image segmentation.Further,we construct the discriminative model and design the superpixel force to improve segmentation effects.In conclusion,the main work of this paper is progressively as follows:(1)Based on organic integration of the Gaussian mixture model and active contour model,a new method is proposed.Experimental results show that the method has inherited advantages of both the CV model and the Gaussian Mixture Model,thus shows a favorable performance in seg-mentation effects.(2)Based on above work,we indroduce the user-interaction-gener-ated lines marked with foreground and background as supervised information,and use semi-supervised generative method to improve the new method.Experimental results indicate the new method can effectively avoid ambiguous segmentation effects and improving user satisfaction of image segmentation.(3)Based on the second work,superpixel force was introduced into the active contour model.Firstly,we generate superpixels.Based on this,a discriminative model is established to design the superpixel force,so that a superpixel force data item is constructed as one of the evolution data item in the active contour model,and a new method is proposed.Experimental results show that the method can further improve effects of image segmentation.