Research On Image And Video Segmentation Methods Based On Variational Model And Graph Cuts Optimization

Image segmentation is to divide image into different regions, and each region has its specific meaning. As one of the difficult problems in computer vision, object segmentation for image and video has been a quite challenging research subject. Image and video segmentation supplies nearly the most basic features for the applications which are based on image and video analysis, and the quality of the segmentation results has a direct effect on the validity of subsequent task in image and video analysis. Recently, those image segmentation methods based on variational model have received widespread concern, and there have appeared a large number of new theories and methods which have been proved to be effective in practical application. This kind of segmentation methods combines the theoretical basis of the geometry, physics and approximation theory, and integrates the feature information of the image and the priori knowledge of the target size, shape and location by minimizing an energy function, so as to segment the objects from background effectively. At the same time, the graph cuts theory becomes a new hotspot of the image segmentation field for its good characteristics.To address the above problems in the field of image and video segmentation, this work provides a good study about the traditional methods of image segmentation, especially the variational model and the Graph Cuts optimization based image segmentation methods. Additionally, a new variational model is proposed, and the Graph cuts optimization framework is optimally redesigned, to efficiently segment objects in images and videos. In details, the main innovative works are described in the following.Firstly, to address the issues that CV model could hardly segment heterogeneous objects or multiple objects with different intensity distributions which often occur in real applications, this work proposes a multiple piecewise constant with geodesic active contour (MPC-GAC) model. The new model generalizes the region-based active contour model by Chan and Vese and merges the edge-based active contour by Caselles et al. to inherit the advantages of regions based as well edge based image segmentation models. We show that the new MPC-GAC energy functional can be iteratively minimized by graph cuts algorithms with high computational efficiency compared with the level set framework. This iterative algorithm alternates between the piecewise constant functional learning and the foreground and background updating so that the energy value gradually decreases to the minimum of the energy functional. Secondly, an iteratively unsupervised image segmentation algorithm is developed, which is based on our proposed multiphase multiple piecewise constant (MMPC) model and its graph cuts optimization. The MMPC model use multiple constants to models each phase instead of one single constant used in Chan and Vese (CV) model and cartoon limit so that heterogeneous image object segmentation can be effectively deal with. We show that the multiphase optimization problem based on our proposed model can be approximately solved by graph cuts methods. The Four-Color theorem is used to relabel the regions of image after every iteration, which makes it possible to represent and segment an arbitrary number of regions in image with only four phases. Therefore, the computational cost and memory usage are greatly reduced.Finally, a new dynamic object extraction approach is proposed, which is based on multiple Gaussian background modeling and graph cuts optimization of variational model. The dynamic objects are first extracted by modeling each pixel intensity as a mixture of Gaussians. Due to the lack of the spatial dependencies of neighboring pixel colors that may be caused by a variety of real dynamic motion, the performance of the multiple Gaussian background modeling method often notably deteriorates when dynamic textures do not repeat exactly. To overcome this problem, we present a variational model and consider dynamic object extraction as an energy minimization problem. The parameters in the model are learned from the result of multiple Gaussian background modeling. We then show that the presented variational model can be efficiently discretely optimized by the graph cuts method. Therefore, the spatial dependent information of neighboring pixel can be effectively integrated to obtain better dynamic object extraction results.In this dissertation, a large number of comparing experiments demonstrate the superior performance of the proposed methods in this work.