Research On Image Segmentation And Object Tracking Algorithm Based On Level Set Theory

Image segmentation and object tracking are two basic and key technologies in the field of computer vision.They play an important role in the practical application of real life.Among them,as a pre-processing step of image processing,the segmentation results of image segmentation directly affect the follow-up operation.Object tracking,another important branch of computer vision,plays an important role in video compression,traffic control,remote sensing monitoring and medical assistance.At present,a large number of image segmentation and object tracking algorithms have been applied in the field of computer vision.Among them,as an extension of active contour model,level set method has become a very popular segmentation and tracking algorithm since its remarkable advantages: flexible topology and stable numerical calculation.Although the theory of level set is very mature,the accuracy of segmentation and tracking still needs to be improved due to the influence of image intensity inhomogeneity,noise,complex background,object occlusion,object blurring and non-rigid change of object.At present,how to propose a level set model that can solve the above problems quickly,accurately and robustly,has become the focus and difficulty of the level set field,which is also an urgent problem to be solved at this stage.So,this paper proposes a series of effective segmentation and tracking algorithms based on level set theory.Specifically,the main research contents and innovations of this paper include:(1)An adaptive scale level set algorithm is proposed to deal with the inhomogeneity of image intensity.Firstly,in order to alleviate the influence of inhomogeneity on segmentation results,a bias field estimation model is established to estimate and correct the inhomogeneity,and it is integrated into the level set model to simultaneously achieve image segmentation and intensity inhomogeneity correction.Then,based on the image information entropy,an adaptive scale operator is defined,which can adjust the scale of the algorithm adaptively according to the inhomogeneity degree of image intensity and thus improve the segmentation speed and robustness of the algorithm.Finally,comparison experiments with state-of-the-art level set model validate the superiority of the proposed algorithm in accuracy for segmenting images with severe intensity inhomogeneity.(2)Aiming at the problem of image noise pollution,a hybrid level set algorithm based on local and global information is proposed.Specifically,by incorporating the global image information into the relative entropy loss function,the global energy term of the algorithm is constructed to guide the evolution curve to converge quickly to the target boundary.Secondly,by incorporating the bias field estimation term with level set model,the local energy term of the algorithm is constructed to extract the target contour accurately.Unlike the traditional hybrid level set algorithm,which uses fixed values to adjust the weights between local and global items,an adaptive scale operator is defined in this paper to realize the adaptive adjustment of the weights between local and global items.The segmentation results for images with different kinds and degrees of noise show that the proposed algorithm is superior to most level set algorithms in terms of segmentation speed and robustness to noise.(3)In order to improve the performance of existing level set algorithms on color images,a saliency supervised level set algorithm is proposed,which makes up for the defect of existing level set algorithms on segmenting color images and improves the application rate of level set algorithm in real life.Firstly,the saliency detection algorithm is used to extract the image saliency as the target score map.By integrating the target score map into the level set model,the global energy term of the algorithm is defined to supervise curve evolution of level set model.Then,based on CIEL*a*b* color space information,the local energy term of the algorithm is defined to extract object boundary in detail.In addition,in light of most level set algorithms use manual method to obtain the initial contour of the target,which is not only time-consuming but also complex,to this end,in this paper,we propose an automatic initialization method to automatically obtain the initial contour of the algorithm,which improves the intelligence and segmentation efficiency of the algorithm.Finally,comparison results with classical and popular level set algorithms validate that the proposed algorithm improves the performance of existing level set algorithms on color images.Comparison experiments with saliency detection algorithm on five datasets validate the superiority of the proposed algorithm in segmentation accuracy.(4)In order to alleviate the contamination of video background pixels on target appearance model,a level set algorithm guided by support vector machine(SVM)for object contour tracking is proposed,which realizes the accurate tracking of object contour and improves the accuracy of target tracking.Firstly,an appearance model of current frame is obtained by repeatedly training and relabeling the original target and background label using a competing one-class of support vector machines(COSVM).Then,by integrating the trained appearance model,an edge detection operator and the spatial information of video frame into the level set model,we propose our algorithm,which can accurately extract object boundary and feed back the results to the support vector machine to update the appearance model.Similarly,we can complete the object extraction for next frame.When target occlusion,blurring and non-rigid deformation occur in next frame,the particle filter algorithm is introduced to train the appearance model that is incorporated into the level set model to accurately estimate the position of the occluded target and avoid the loss of the target.Finally,the tracking results on three open datasets validate the feasibility and effectiveness of the proposed algorithm.