Research On Active Contour Models For Complicated Image Segmentation

The active contour method for image segmentation was first proposed by M.Kass et al.in 1987.Over the last decades,it has been a research hotspot in the field of image processing.With research development,numerous active contour models were proposed and widely applied in military,medicine and so on.However,there still exist some problems of the state of art models associated with poor capability of segmenting complicated images,for not only the edge-based but also the region-based ones.In more detail,when an object to be segmented has a complicated geometry,edge-based models trend to give an under-segmentation,because such kind of models just use the local information of image.On the other hand,since region-based models are constructed by using the global information of image,the recovery of one object of interest may be influenced by other objects contained in the background.As a result,when the image to be segmented has a complicated feature(such as intensity and texture)distribution,a false segmentation would be given.These problems greatly restrain the wider application of active contours.In this paper,several active contour models for complicated image segmentation are proposed,in order to enrich the research and extend the application of active contours.Firstly,for solving the problems of active contours based on gradient vector flow(GVF)associated with difficulties in evolving over the saddle points and stationary points of GVF field,an active contour model named with "active contours driven by divergence of gradient vector flow" is proposed.The non-conservation of GVF field is deeply investigated and an important characteristic of the divergence of GVF field is concluded.Then an energy functional is constructed and its gradient flow(negative gradient)is used as the external force of new model for active contour evolution.The new model can drive active contours over the saddle points and stationary point of GVF field,and so has the capability of recovering objects with complex geometries.While the new model holds several features of the GVF method,such as good robustness to noise and a wider capture range.Secondly,borrowing from the differential geometry viewpoint a new edge map operator is proposed,and the computation of the edge feature of scalar and vector-valued images is unified.Using the new edge map operator,many edge-based models are extended to the case of vector-valued image segmentation.Especially,a method of segmenting color images by using active contours driven by divergence of GVF is proposed.Thirdly,an active contour model named with "active contours driven by the gradient flow of minimum probability density integration(MPDI)" is proposed,for solving the problem of the existing nonparametric statistical models associated with giving false segmentation to images with complicated feature(such as intensity and texture)distributions.The density functions of the intensity distributions inside and outside the active contour is first computed.Following this,an energy functional of minimum probability density integration(MPDI)is constructed,and whose gradient flow(negative gradient)is used as the external force of new model for active contour evolution.The new active contour model can availably segment images with complicated feature distributions,and requires a clear initial setting.