Research On Image Segmentation Based On Active Contour Model

Image segmentation is the key step for image understanding and analysis. Its main aim is to separate the whole image into several sub-regions which share one or more similar statistical properties. Image segmentation always restricts the development of high level image processing, however, in recent years the image segmentation method based on active contour model has received extensive attention theoretically and technologically by the image processing experts, because active contour model has very rigid mathematical base and efficient numerical scheme, in addition it is very promising to overcome some shortcomings of traditional image segmentation methods such as breakpoints on final contour etc.In this paper, we first gived a comprensive summary of present image segmentation methods and we surveyed the development of active contour model in detail. Then we introduced several famous active contour models. Two new active contour models were proposed based on the problems of the state-of-the-art research of active contour model.(1) A new active contour model based on global-local image region information. The aim of the new model is to solve the problems of segmenting low contrast, intensity inhomogenerity and noise pollution images. We first analyzed the advantages and disadvantages of CV model and LBF model, then we obtained the new model by proposing a new global penalty function, which improves the combination of CV model and LBF model. The new model was applied on synthetic and real images. The experimental results show that the new model can segment the object correctly from images with low contrast, intensity inhomgenerity and strong noise.(2) Liver CT-slice image segmentation by active contour model based on prior shape. Applying high level image information to solve the lower-layer problems is one of the best way to deal with such situation that objects are occluded partly and subject to strong noise. In this paper, based on the fact that adjacent liver slices are continuous spatially, the mean liver shape in a series of liver slices can act as prior shape naturally. So, we first segmented the liver slices to obtain the mean initial liver region by applying double-thresholds and mathematical morphological filtering. Then a new signed distance function was proposed on the mean initial liver region and was embed to active contour model's energy functional by the difference of shape between the current image's level set function and the prior shape's. The experimental results show that our new model can successfully segment liver from the series of liver slices and occluded liver images.Finally, we pointed out the direction of further research.