Efficient Kernel Density Estimation of Shape and Intensity Priors for Level Set Segmentation

Abstract

We propose a nonlinear statistical shape model for level set segmentation that can be efficiently implemented. Given a set of training shapes, we perform a kernel density estimation in the low-dimensional subspace spanned by the training shapes. In this way, we are able to combine an accurate model of the statistical shape distribution with efficient optimization in a finite-dimensional subspace. In a Bayesian inference framework, we integrate the nonlinear shape model with a nonparametric intensity model and a set of pose parameters that are estimated in a more direct data-driven manner than in previously proposed level set methods. Quantitative results show superior performance (regarding runtime and segmentation accuracy) of the proposed nonparametric shape prior over existing approaches.

Shape-Driven 3D Segmentation Using Spherical Wavelets

“…Furthermore, we see that as the resolution level is increased for the α parameters, the Mscale segmentation is able to capture finer details. We then validated the full segmentation algorithm, using the proposed image force in Equation 5. The results of the validation for both algorithms are shown in Table 1.…”

“…With region-based energies, the force that influences the evolution of a contour depends on more global statistical information [4,5]. We employ the discrete version of a segmentation energy presented in [5]:…”

“…Object segmentation with deformable models and statistical shape modelling are often combined to obtain a robust and accurate segmentation [2,3,4,5]. Active shape models (ASMs) [3] are a standard technique for statistical segmentation tasks based on a prior learned over a point distribution models (PDM): landmarks on the shape are used as parameters and a joint prior probability distribution is learned using principal component analysis (PCA) over the landmarks.…”

TouchCut: Fast image and video segmentation using single-touch interaction

“…Tsai et al [62] proposed a very efficient implementation of shape-driven level set segmentation by directly optimizing in the linear subspace spanned by the principal components. The use of nonparametric density estimation to model larger classes of level set based shape distributions was proposed by Cremers et al [18] and Rousson and Cremers [58]. Paragios [50] proposes to integrate user interactive constraints to level set framework by taking into account distance from seeds, which is similar in aim to the geometric prior proposed in our work.…”

“…Higher level prior knowledge such as geometric shape priors have been introduced to level set framework [59,53,21,9,62,58,18,20]. Variational integration of the shape prior based on the assumption of a Gaussian distribution were proposed by Rousson and Paragios [59].…”