Image Segmentation Models And Applications Based On Global And Local Information And The Split Bregman Method

Image segmentation is accepted as a basic task and an important part of image processing,and also a basic preparatory step in the field of computer vision.There are many existing segmentation methods,such as edge detection,threshold segmentation and regional growth.Among all those methods,the most common and successful type is active contour model.Although the traditional active contour models have achieved good results,there are still some problems to be solved: unable to handle inhomogeneous images,sensitive to initial contours,not robust to noise,slow to converge and so on.Moreover,these models are non-convex,which can lead to local minimal solutions in the process of minimization.In order to solve the problem of non-convexity,we introduce the idea of global convex segmentation method into new segmentation models.These classical active contour models minimize the energy functionals by applying the gradient descent method,but the gradient descent method converges slowly.Therefore,we introduce the Split Bregman method to the minimization.Then we can get faster segmentation models.For brain magnetic resonance(MR)images,with an improved active contour model combining local and global information dynamically,brain contours are obtained and feature extractions of images are carried out.For real,synthetic and medical images,in addition to considering global and local information,with the global convex segmentation method and the split Bregman method,neighborhood information of global information is also considered.For medical MR images,considering the bias field information,segmentation results are obtained while images are corrected.This paper introduces the following segmentation models.An improved active contour model combining local and global information dynamically(GCLGIF)has successfully segmented some real and synthetic images with inhomogeneous intensity,and compared with traditional active contour models,this model has been improved in accuracy,robustness and rapidity.Therefore,we further organize and analyze the model,in order to segment brain contours and lesions at the same time,we use the multi-phase model to segment brain MR images.Multi-phase segmentation of brain MR images can not only get contours of brains,but also get contours of lesions.Based on global information and local information,we introduce the neighborhood information of global information and the edge detection function,and consider the global convex segmentation method and the split Bregman method,we propose a new segmentation model based on local and global and neighborhood information(NLGIF).Considering many factors above of all,an energy functional is defined and the energy functional is minimized by using the split Bregman method.The NLGIF model can segment more general images with inhomogeneous intensity,which can not only improve the accuracy of segmentation and the robustness to the selection of initial contours and noise,but also can segment images faster.Therefore,the NLGIF model is an efficient and robust segmentation model.Based on the bias field information and the split Bregman method,an improved active contour model incorporating bias field correction(BFC)is proposed for segmentation of medical MR images.The BFC model can not only get the segmentation results,but also can get bias field correction images.When the energy functional is defined,the bias field information is included,and we apply the split Bregman method to the minimization of the energy functional.Numerical experiments are performed and we compare the BFC model with other models.Experimental results also show that the new model is robust to the initial contour and noise,and the segmentation is more accurate and faster.In three fast image segmentation models,the GCLGIF model is applied to some brain MR images.Then segmentation results are obtained,and features of images are extracted.The NLGIF model is used to segment synthetic and real images,not only accurate results are obtained,but also the robustness is improved.We apply the BFC model to synthetic and MR images,and segmentation results are obtained while the bias field correction images are obtained.The above models have achieved good segmentation results,and the effectiveness of these models has been proved.In addition,we compare experimental results with other models to demonstrate the superiority of these new models,which turns out more accurate segmentation results and less sensitive performance to noise and initial contours.The application of the split Bregman method guarantees the fast convergence of the algorithm,thus saves time of segmentation.