Research On Local Region Fitting-based Active Contour Model For Image Segmentation

Image segmentation is an important and complex task in the field of image processing and computer vision.It plays an important role in automatic driving,3D reconstruction and medical image analysis.In recent years,the image segmentation method based on active contour model has developed rapidly due to the advantages that it can obtain the sub-pixel accuracy of target boundary and provide smooth closed contour as segmentation results,which is contribute to further image analysis and recognition.This paper studies the active contour model based on local region fitting energy.The traditional local fitting model can effectively segment images with intensity inhomogeneity,but their segmentation speed is slow and sensitive to the initial contour.To solve these problems,three improved models are proposed in this paper:1.The model driven by local fitting energy with the same direction.The core of this method is to exchange the fitting values in the local region where the curve evolution direction is opposite,so that the fitting value inside the contour is always larger(or smaller)than that outside the contour in the process of curve evolution.Therefore,the whole curve will evolve continuously along the inner boundary(or outer boundary)of the object,and will not stay in the interior of the object,namely it can avoid the local optimal solution when minimizing the energy.This method retains the advantages of the traditional local fitting model and greatly improves the robustness of the initial contour.2.The model combined local region fitting energy and optimized Lo G energy.First,an energy functional is proposed to optimize the Lo G edge operator,which can smooth the homogeneous regions and meanwhile enhance edge information.Then,the optimized Lo G energy is linearly combined with the traditional local region fitting energy which helps the curve towards the object boundaries by using local image intensities.Compared with the traditional local fitting model,the proposed model not only has good robustness to the initial contour,but also has faster segmentation speed.3.The model driven by local pre-fitting energy.The core idea of this method is to calculate the average intensities of the local image before the curve evolution,namely define two local pre-fitting functions.Compared with the traditional fitting functions,the pre-fitting functions are independent to the level set function,and they do not need to be updated in each iteration.Therefore,the model has lower computation cost and faster segmentation speed.In addition,the model has strong robustness to the initial contour,which allows the initial level set function to be a constant function.Furthermore,according to the final fitting image,the proposed model can also be applied to image denoising and image contrast improvement.The experimental results have showed that above three improved methods can effectively segment the images with intensity inhomogeneity and have a good segmentation results for the images with weak edges and noises.Compared with the traditional local fitting model,the proposed first model overcomes the sensitivity to the initial contour,and the second model has faster segmentation speed and stronger robustness to the initial contour,and the third model significantly improves segmentation efficiency and the robustness of initial contour.