Search

We present in this chapter an overview of the Mumford and Shah model for image segmentation. We discuss its various formulations, some of its properties, the mathematical framework, and several approximations. We also present numerical algorithms and segmentation results using the Ambrosio–Tortorelli phase-field approximations on one hand, and using the level set formulations on the other hand. Several applications of the Mumford–Shah problem to image restoration are also presented. [ABSTRACT FROM AUTHOR]

We wish to recover an image corrupted by blur and Gaussian or impulse noise, in a variational framework. We use two data-fidelity terms depending on the noise, and several local and nonlocal regularizers. Inspired by Buades-Coll-Morel, Gilboa-Osher, and other nonlocal models, we propose nonlocal versions of the Ambrosio-Tortorelli and Shah approximations to Mumford-Shah-like regularizing functionals, with applications to image deblurring in the presence of noise. In the case of impulse noise model, we propose a necessary preprocessing step for the computation of the weight function. Experimental results show that these nonlocal MS regularizers yield better results than the corresponding local ones (proposed for deblurring by Bar et al.) in both noise models; moreover, these perform better than the nonlocal total variation in the presence of impulse noise. Characterization of minimizers is also given. [ABSTRACT FROM AUTHOR]

In this paper, we present a new non-parametric combined segmentation and registration method. The problem is cast as an optimization one, combining a matching criterion based on the active contour without edges [4] for segmentation, and a nonlinear-elasticity-based smoother on the displacement vector field. This modeling is twofold: first, registration is jointly performed with segmentation since guided by the segmentation process; it means that the algorithm produces both a smooth mapping between the two shapes and the segmentation of the object contained in the reference image. Secondly, the use of a nonlinear-elasticity-type regularizer allows large deformations to occur, which makes the model comparable in this point with the viscous fluid registration method [7]. Several applications are proposed to demonstrate the potential of this method to both segmentation of one single image and to registration between two images. [ABSTRACT FROM AUTHOR]

We introduce several color image restoration algorithms based on the Mumford-Shah model and nonlocal image information. The standard Ambrosio-Tortorelli and Shah models are defined to work in a small local neighborhood, which are sufficient to denoise smooth regions with sharp boundaries. However, textures are not local in nature and require semi-local/non-local information to be denoised efficiently. Inspired from recent work (NL-means of Buades, Coll, Morel and NL-TV of Gilboa, Osher), we extend the standard models of Ambrosio-Tortorelli and Shah approximations to Mumford-Shah functionals to work with nonlocal information, for better restoration of fine structures and textures. We present several applications of the proposed nonlocal MS regularizers in image processing such as color image denoising, color image deblurring in the presence of Gaussian or impulse noise, color image inpainting, and color image super-resolution. In the formulation of nonlocal variational models for the image deblurring with impulse noise, we propose an efficient preprocessing step for the computation of the weight function w. In all the applications, the proposed nonlocal regularizers produce superior results over the local ones, especially in image inpainting with large missing regions. Experimental results and comparisons between the proposed nonlocal methods and the local ones are shown. [ABSTRACT FROM AUTHOR]

This paper is devoted to piecewise-constant segmentation of images using a curve evolution approach in a variational formulation. The problem to be solved is also called the minimal partition problem, as formulated by Mumford and Shah [20]. The proposed new models are extensions of the techniques previously introduced in [9], [10], [27]. We represent here the set of boundaries of the segmentation implicitly, by a multilayer of level-lines of a continuous function. In the standard approach of front propagation, only one level line is used to represent the boundary. The multilayer idea is inspired from previous work on island dynamics for epitaxial growth [14], [4]. Using a multilayer level set approach, the computational cost is smaller and in some applications, a nested structure of the level lines can be useful. [ABSTRACT FROM AUTHOR]