Research On Method Of Image Segmentation Based On Geometric Active Contour Model

Image segmentation is a basic and important problem in image processing, and it is also the foundation of researches on image analysis and computer vision. Image segmentation is to divide an image into some disjoint subregions, in order to achieve the purpose of separating target object or region of interest from image region. There are many image segmentation methods now. Recently, the method of image segmentation based on geometric active contour model, as one of the novel and efficient segmentation methods, is gradually turned into a hotspot of research and application.On the basis of fully realizing active contour model and the related background knowledge, this paper mainly studies several classical geometric active contour models based on region information, and it also points out the existing shortcomings in these models. In order to improve their shortcomings, this paper presents the improved LBF model and its image segmentation algorithm by implementing three improvements on the basis of LGIF model. These improvements are respectively:(1) In LGIF model, it uses the LBF model which has lager scale parameter σ instead of C-V model, because this kind of LBF model has not only global characteristics but also local characteristics;(2) It firstly introduces local entropy which is gotten after data processing into LGIF model, then it calculates weight parameter ω automatically. This method overcomes the shortcoming that the calculation of weight parameter in LGIF model needs artificial participation;(3) In order to be beneficial to automatic computer calculations and avoid too much useless cyclic iterations, it presents a new stop criterion.In order to evaluate the validity and practicability of the improved model and algorithm, some experiments are done respectively with artificial images、SAR images and medical images. Results show that our model not only improve LGIF model effectively but also improve the shortcoming of LBF model, and the shortcoming is that LBF model is sensitive to the place of initial contour curve. Meanwhile, by comparing segmentation results between our algorithm and traditional image segmentation algorithms, this paper obtains a conclusion of that active contour model can more satisfy the requirements of image segmentation in medical field and remote sensing field.