A Variational Level Set Based On Adaptive Initialization And Multi-feature Integration

The level set method has been widely used in video image processing,medical research and so on because it easily achieves curve topology change,and has high calculation precision and strong robustness.As the advantages of multi-information fusion,the variational level set method has been widely concerned by researchers in recent years and has become a hot research topic in the level set method.However,the variational level set method cannot adaptively obtain the initialization curve,and solve the problem of detection of multiples targets in complex images.And how to integrate a variety of image information features in the variational level set model to obtain more accurate segmentation results.These problems are the research hotspots and technical difficulties of the variational level set method currently.In this paper,the variational level set methods have been deeply investigated.The main works can be summarized as follows:(1)To solve the problem of adaptive initialization in the variational level set method,the paper proposes a variational level set based on adaptive initialization,and applied to the multi-target detection problem in complex background.In this mothed,the inter-frame difference algorithm is combined with K-means clustering algorithm to obtain multiple initialization curves,and then reduce the noise by morphology method,which can estimate the position and the size of the moving target in complex background.Then the mothed combines the adaptive initialization method to obtain N-1 curves which segment the image to N regions,and each curve represents on region.And then,the varitational level set without re-initialization is extended form single target detection to multiple target detection,and improve the model's ability to deal with the images of nonuniform gray.The experimental results show that the proposed method can accurately locate target contours of different scales and gray levels so as to improve the evolution efficiency and accuracy of the algorithm.(2)This paper proposes a variational level set method based on multi-attribute feature integration,and applied to the superpixel segmentation of SAR image to get good results.The traditional variational level set model can only integrate the single attribute feature of image.To solve this problem,the paper combines the advantages of multiple features in SAR image,including the speckle noise,texture and edge features,and integrated these features into the energy function of the variational level set.And the integrated energy functions of variational level set are applied in Turbopixles algorithm to respectively propose the variational level set of integrated speckle noise for superpixel segmentation(GT-pixles),the variational level set of integrated speckle noise and edge feature for superpixel segmentation(GT-pixles-E),the variational level set of integrated speckle noise and texture feature for superpixel segmentation(GT-pixles-T),and the variational level set of integrated speckle noise,edge feature and texture feature for superpixel segmentation(GT-pixles-E&T).Experiments indicate that the proposed mothed can effectively use the SAR information to overcome the influence of speckle noise and improve the accuracy of super pixels segmentation of SAR image.