Research On Active Contour Model Based On Image Segmentation

Image segmentation is the technique that divides image into similar feature regions,which is the basis for advanced image processing, such as image understanding andrecognition. Image segmentation method based on the C-V active contour model can notonly rely on adequate and reliable low-level image features, but also rely on high-levelknowledge. Compared with the traditional image segmentation method, the C-V activecontour model has strong advantages. However, the model makes well use of the regioninformation, so images containing the heterogeneous and complex background can not besegmented well. In addition, in practical application, for the computing complexity of theC-V model, the real-time problem is also needed to take into account. Hence, furtherresearches on the C-V active contour model are much necessary in both theoretical andapplication aspects.Intensive studies are carried out in this dissertation on the key techniques of the C-Vactive contour model, which include more image information considered in the model andregion division. The contributions of this dissertation are as follows:Firstly, the classical geodesic active contour (GAC) model tends to produce fakeedges in smooth region, while the C-V model cannot effectively detect images with holesand obtain the precise boundary. To address the above issues, this dissertation proposes anadaptive mixture model synthesizing the GAC model and the C-V model by a weightfunction. According to image characteristics, the proposed model can adaptively adjust theweight function. By this way, the model develops the advantage of the GAC model in theregion with rich textures or edges, while exploits the advantage of the C-V model insmooth local region. Moreover, the proposed model is extended to the vector-valuedimages. By experiments, it is verified that the proposed model obtains better results thanthe traditional models. Secondly, an improved variational level set for image segmentation framework thatcombines the edge information and region information is proposed. The framework isimplemented on the C-V active contour, which is a typical region-based active contourmodel. We also incorporate the edge information into the C-V active contour model tomake the evolving curve moving on the object boundary. To demonstrate the versatility ofthis framework, the proposed framework uses three different edge functions: the simpleGradient Vector function, the normal Gradient Modulus function, and the WaveletModulus function. Experimental results show that the proposed framework can segmentthe noisy blurry boundaries and intensity heterogeneous images. Finally, we show resultson challenging images to illustrate the accurate segmentations by our proposed framework.Thirdly, an unsupervised image segmentation technique is proposed. Firstly, forobtaining a multiresolution representation of the segmented image, the probability modelof the nonsubsampled contourlet coefficients of the image is established. A region-basedactive contour model is then applied to the multiresolution representation for segmentingthe image. The proposed technique has been conducted on challenging images to illustratethe robust and accurate segmentations. At last, an in-depth study of the behaviors of theabove techniques in response to the proposed model is given, and the segmentation resultsare compared with several state-of-the-art methods.At last, we propose a new multi-object segmentation method based on C-V model,which is a level set method for segmentation on high-noise and blurred edge images. Inorder to extend the C-V model to the multi-object segmentation, in this dissertation, regiondivision procedure is introduced. In the procedure, image domain is firstly divided intomultiple sub-regions, and then curve evolving is performed in each sub-region. In addition,region merging between the regions during evolving is considered, and necessary theoriesare analyzed mathematically. Theoretical analysis and computer simulation data illustratethat the proposed method can effectively segment images with multiple different meanobjects, and the segmentation process is sped up.The first two parts in this dissertation are the improved work of combining the imageboundary information and the image area information; while the the third part mainlyconsiders the frequency domain of the image information, and combines the imagefrequency domain and the spatial information. Image segmentation process better uses theimage information; the fourth part applies the active contour model method to multi-objective image segmentation process, and the algorithm can be further applied totraffic video segmentation and other practical image segmentation applications.