Research On Novel Extraction Methods For Object Edges And Contours

Recently,with the rapid development of the computer science and technology,theimage processing and machine vision have attracted many attentions,which are applied ina variety of corresponding fields.As one of key technologies in image processing,imagesegmentation is a base of image understanding,image analysis and image recognition.Although there are a lot of works addressing how to extract objects edges and contours,itis still a challenging problem in image processing.According to rigorous mathematicstheory,reliable physics principle and practical applications,the novel theories andapproaches to extracting object edges and contours are studied in this thesis.The maincontents of this thesis include:Firstly,by using inherent relationship between region integral and curve integral,designing the image transform operator,a novel snake model using both boundary andregion information was proposed,which more directly integrated the prior knowledgeincluding region information with the traditional deformable model,so that the abilityabout evolving contour curve was improved.Experimental results demonstrated ourmethod not only extended the initialization of contour,alleviated the sensitivity to imagenoise,but also improved the capacity to converge into complex boundary.As can be seenfrom analysis of this thesis,the including of priori-information of object regions is perfectcoherence with the deformable models not only in mathematics,but also in physicsmeaning,which provide a valuable reference for the introduction of priori-information ofobject regions.Secondly,based on the physical and motional characteristics of quantum particle,anovel model,namely quantum statistical deformable model (QSDM),motivated quantummechanics for image edge detection was presented.Then,its convergence ability wasproved.The error about its discretized formation was estimated and its parametersensitivity was analyzed,such that a more systematic model was obtained.This bridges the gap between quantum mechanics and objects edges extraction,which forms a noveltheory about objects edges extraction.Experimental results demonstrated our method hasnot only the capability to makes edge extraction with arbitrary initialization and topologychanges,but also improved the capacity to remove image noise and converge into thedeeply thin concavity.Furthermore,the potential of proposed method was demonstratedby the analysis of the complexity and parameter sensitivity.Subsequently,in order to improve the independence between neighboring samples inthe sample set and the convergence efficiency,the sampling method in the QSDM wasmodified by using multilevel Metropolis sampling (MMS),which leads to an efficientsampling.Therefore,the convergence ability and precision of numerical computation wereimproved.Moreover,the stability of the proposed method was ensured by the analysis ofconvergence and error estimation.Experimental results demonstrated the efficiency andpracticability were improved by the introduction of MMS sampling in the QSDM.Finally,according to of the characteristic of densely scattered points from the QSDM,a novel curve reconstruction method based on the A* method was proposed.It can furtherdecrease the influence of noise points on the reconstructed curves in the proposed method,and the satisfying reconstructed curves from the densely scattered points with differenttopology forms were obtained.Experimental results demonstrated the satisfying curvescan be reconstructed in densely scattered points from QSDM by using the proposedmethod.