Completion Of Illusory Contours And Missing Boundaries

As an important branch of image processing, image segmentation has been developing rapidly. And at the same time, as a special image segmentation technique, capture of illusory contours has been paid more and more attention. Image segmentation is an important image analysis technique, and is a technique and process to segment the image into several characteristic parts and to pick up the interesting objects. So-called illusory contours are closed in vision but unclosed in real. Some parts of illusory contour are made up of existing objects'real boundaries, and the others are boundaries which do not really exist but are caused by visual system. Kanizsa triangle and square are common illusory contours.Some important theoretic technologies about image segmentation are studied, by which capturing illusory contours models are analyzed and improved. Firstly, several image segmentation models are analyzed, especially for LXDF model based on level set method without re-initialization which is proposed by Li C. et al. The new model is obtained from LXDF model by revising the edge indicator function introduced by Li C. et al. And then the improved model can segment the images in which the objects have sharp corners. The advantage of the improved model is showed by numerical experiments. Secondly, several models of capturing illusory contours are analyzed. Zhu-Chan model using curvature information, proposed by Zhu W. et al., is paid more attention. The constrained information of signed distance function from LXDF model is introduced to Zhu-Chan model, and then an improved Zhu-Chan model is obtained. By the theory of segmenting first and then capturing being proposed, the failure of pretreatment to images with common scale values by Hamilton-Jacobi equation in original Zhu-Chan model is eliminated, from which Zhu-Chan model is extended. The numerical experiments show the advantages of the improved model.