Research On Medical Image Segmentation Based On Geometric Deformable Model

Based on the thorough understanding and research of Level Set method and Chan-Vese Model, this paper proposes the following improved methods through a lot of experiments and analyses in order to improve and solve the problems such as the speed of segmentation and the segmentation precision of weak boundaries.First of all, a novel segmentation model based on exponential boundary gradient speeding term is proposed, which can not only selectively speed the segmentation of specific objects, but also can improve the segmentation accuracy of objects having weak boundaries. Large numbers of experiments indicate this model has better segmentation efficiency for specific objects, compared to the gradient advanced segmentation model. What's more, the quotation of internal energy term eliminates the time-consuming re-initialization process.Secondly, an improved unsupervised hierarchical segmentation model is suggested which utilizes the concept of entropy correlative coefficient(ECC) in the mutual information field. Through the control of segmentation process using both ECC and Var, this method, compared to the traditional Chan-Vese Model, can remarkably improve the precision of objects having weak boundaries with the same Level Set function by way of one-by-one class.Finally, a simple Dual Level Set model is proposed based on the Chan-Vese Model and the idea of parallel computation. This model can respectively speed up the evolvement of the two Level Set curves to two objects having different boundary gradient value at the same time in the same image. Moreover, in order to solve the problem that the segmentation speed of Chan-Vese Model is sensitive to the position of initial Level Set curve, by making use of this model, two Level Set curves can be placed at the neighborhood of two different regions of interest. When the time reaches the threshold value, the Level Set curve having less energy value is selected as the final segmentation result.