Research On Image Segmentation Based On Active Contour Model

Image segmentation is an important question in the field of computer vision. Inrecent years, image segmentation based on active contour model becomes a moreactive research hotspot in the image segmentation field. Active contour model has atypical characteristic, which concentrates the image data, the initial contour selection,target feature and motion in a variational framework, and it will transform theproblem of looking for the boundary of the object into the problem of solvingminimum of the energy functional. At the same time, it has rigorous mathematicaltheory, subpixel accuracy and high efficient numerical calculation.This paper accomplished innovation to traditional classical active contour modelby reading numerous domestic and foreign related literatures and doing someexperiments to further analysis the advantages and disadvantages of classical activecontour model. My work can provide some new methods and ideas for obtaining moregeneral image segmentation method, while both theoretical analysis and simulationresults proved that two new models proposed in paper have obvious advantagescompared with the traditional classic active contour model, and have certain researchvalue.The main innovation points: Combining Local Binary Fitting (LBF) model andChan-Vese (CV) model, a new model which named LGIF model is proposed. It useslocal and global information effectively for image segmentation. The new model cansolve the problem such as LBF model easily falling into local minimum and errorsegmentation; moreover it can also solve the evolution speed too slow by CV model.LGIF model inherits the advantages of local information and global informationactive contour model. Experiments show our model can be used to segment imageswith intensity inhomogeneous while CV model and LBF model can not; LGIF modelis not sensitive to the initial contour, with strong noise immunity. Comparisons withLBF and CV, the advantages of our model in terms of computational are testified.Combining the Geodesic Active Contour (GAC) model and Distance RegularizedLevel Set Evolution, another new active contour model which named Combining edgeand regional information is proposed. The new model is a combination between edgemodel and region model, and it possesses peculiarities of edge model and region.Finally, I present various experimental results and in particular some examples forCVI, GAC, and CV are not applicable, while the experiments show that my model hasa stronger noise immunity and high efficiency.