Research On Image Segmentation Based On Level Set Regularization

Image segmentation is an important branch of image processing,and it is also an indispensable part of computer vision.Segmentation is the primary operation for in-depth research on images or videos and it plays an important role in the research of target tracking,target recognition,image registration and so on.The image segmentation algorithm based on level set method is an active research direction in recent years.The primary advantage is that it can deal with topology changes effectively.Moreover,the energy model based on level set can be easily combined with other methods,which is widely concerned by researchers.The current level set theory is still developing and changing,and the model itself still has many defects that require further research and improvement.Therefore,this paper carries out the research on the image segmentation technology based on the level set and active contour model.Some research results have been achieved:1.The basic theory of the active contour model is summarized,and the different representations of the contour are introduced,including parameter-based and geometric-based active contours.The advantages and disadvantages of each method are briefly described,and some classical models are introduced.Then,some problems of the model based on the geometric representation(level set method)are analyzed: On the one hand,the level set function is sensitive to initialization,so it is necessary to set up a reasonable initial contour;On the other hand,the stability of the level set function in the iterative process needs to be treated with regularization.We also outlines some of the current solutions to this problem and the shortcomings of these solutions.2.Aiming at the problem that the traditional active contour model is sensitive to initialization,a new initialization scheme based on the saliency detection method is proposed.Due to the non convexity of the energy model,the initial contour needs to be set reasonably.However,when the initial position deviates from the main target,the energy model may easily fall into a local minimum,resulting in an incorrect segmentation result.Based on the saliency map of the image,the position information of the target can be obtained.The modified saliency detection results are used as the initial level set function to automatically generate initial contours for different images,which can effectively prevent the initial position of the contour from deviating from the main target.It avoids the local minimization result,reduces the manual operation process and improves the efficiency of the algorithm.3.For the regularization problem of the level set function,a new regularization method based on Bessel filter is proposed.For the stability of the level set iteration,the traditional model needs to periodically re-initialize the level set function,which is very time-consuming.In recent years,reinitialization-free models have arisen,and the level set function is automatically adjusted by adding regularization item to the energy model.Based on the classical model,the Bessel filter is added to the regularization process to maintain the stability of the level set function in the process of evolution.In addition,based on the advantages of Bessel filter in edge detection,a new edge stop function is proposed.Combined with the traditional model,the evolution curve can be more accurately stopped at the edge position,and the accuracy of the model is improved.Based on the traditional active contour model,this paper improves the algorithm based on the level set function.On the standard data set,the classical model is compared and quantified with the proposed and the simulation results show the effectiveness of the proposed algorithm.At the same time,it is found that the algorithm has some limitations in the experiment,and it needs to be further studied in the future.