Journal of Visual Communication and Image Representation
Reduced set density estimator for object segmentation based on shape probabilistic representation
Highlights
► A nonparametric statistical shape model for object segmentation is proposed. ► PCA on shape probabilistic representation needs less number of eigenmodes than SDF. ► RSDE can provide a high-accuracy and low-cost density estimator.
Introduction
Object segmentation is a fundamental task in image processing and computer vision. Its essential goal is to extract desired objects from the given images. Since the object and background may exhibit very similar intensity characteristics in numerous real-world applications, it is normally not enough to only use the low-level information of the images, such as intensity, color or texture for segmentation, especially when misleading information due to occlusion, clutter and noises exist in the input images. This naturally leads to a need for integrating prior knowledge such as shape information into the segmentation process in order to improve segmentation results. In this paper, by assuming that prior knowledge given by a set of training shapes of expected objects, we focus on the problem of how to exploit such shape priors for object segmentation.
Level set methods were introduced by Osher and Sethian [1]. Since such methods allow implicit representation of the evolving object boundary and automatic changes of its topology, level set methods have become increasingly popular for image segmentation [2], [3]. Recently, to segment images of low quality or with missing data, level set based variational approaches have gained significant attention toward the integration of shape prior into the image segmentation processes [4], [5], [6], [7], [8], [9], [10], [11], [12], [13]. Almost all these works can be considered as a linear combination of two terms: a data-driven term and a shape constraint term. Geometric active contours model [14] and Chan-Vese’s model [15] have become two popular data-driven terms to guide the motion of the active contour. There are two ways to define the shape constraint term. One is commonly defined by an explicit dissimilarity measure between the evolving contour and a given prior contour, and the other is to estimate a statistical distribution from training shapes to guide the evolving contour to the most likely shape of the estimated distribution. Given a set of training shapes, one may impose simple or more complicated distribution functions such as uniform distribution [7], Gaussian distribution [16], or non-parametric estimator [17] to improve segmentation results in the presence of noise or occlusion. In applications, the distribution of training shapes is generally not uniform distribution or Gaussian distribution due to a large variability of shape. Kernel density estimation (KDE) is an efficient approach to model nonlinear distributions of training shapes [12], [13]. In this technique, the density function is estimated by a sum of kernel functions. The kernel number is equal to the size of the training data. When the training data set is very large, the KDE suffers from high computational cost and becomes intractable for subsequent use (e.g., in a real-time applications). Reduced set density estimator (RSDE) was proposed by Girolami and He [18] to solve the above problem by providing a kernel density estimator which employs a small subset of the available data sample to provide similar levels of performance.
Shape is represented implicitly by signed distance function (SDF), and can be easily integrated into level set variational methods as a shape constraint term. These representations have gained much popularity in recent years [4], [5], [6], [7], [8], [9], [10], [11], [12], [13]. The idea is to represent the shape contour C by embedding it in a higher dimension level set functional ϕ, as follows:where Dist(x, C) denotes the Euclidean distance from x to the closet point on C, and out(C) and in(C) represent the regions outside and inside of the contour C, respectively. The contour C can be reconstructed from such representation by taking its zero level set C = {x|ϕ(x) = 0}. Hence, any shape in the plane corresponds to a unique SDF. This shape representation is consistent with the level set framework, and has its advantages since parameterization free and easy handling of topological changes. However, the use of principal component analysis (PCA) on a set of SDF embedding a set of sample shapes has two drawbacks:
- 1.
The space of SDF is not a linear space, e.g., the mean shape and linear combinations of sample shapes are typically no longer SDF. Most existing works only consider very similar shape priors.
- 2.
While the first few principal components are used to capture the most variation on the space of SDF, they will not necessarily capture the variation on the space of the embedded shape contours. Therefore, in contrast to PCA on explicit shape contours, PCA on SDF need to include a larger number of eigenmodes in order to capture certain details of the sample shapes.
Recently, there has been significant research exploring methods to solve these non-convex problems by using convex relaxation methods [19], [20], [21], [22]. In [20], Cremers et al. proposed a shape probabilistic representation (SPR) by relaxing the binary constraint and allowing the binary function to take on values in the interval [0, 1], defined as a mappingthat assigns to every pixel x of the shape domain the probability that this pixel is inside the given shape. In traditional definition of shape, pixels are part of the shape, and only take values 1 (members) or 0 (non-members). It can be described as q:Ω → {0, 1}. Based on the probabilistic definitions, it is easy to get the shape region of the object and the background of image by selecting a τ ϵ [0, 1]. In the experiment, τ is chosen as 0.5. It was shown that the space Q of all probabilistic shapes forms a convex set, and the space spanned by a few training shapes χ = {q1, q2, ⋯qN} forms a convex subset. Arbitrary convex combinations of the set again correspond to a valid shape. For example, the mean is a function which assigns to each point x ϵ Ω the average of all probabilities (Fig. 1). This shape probabilistic representation leads to convex segmentation functional on convex shape spaces.
In this paper, we are building up on the above developments and propose two contributions in order to overcome the discussed limitations:
- 1.
We use probabilistic representation to model the shape prior, and then compute PCA on shape probabilistic representation to capture the variation of the training shapes. In contrast to the commonly used signed distance functions, PCA on shape probabilistic representation needs less number of eigenmodes to capture certain details of the training shapes.
- 2.
RSDE is used to learn the shape prior information in the low-dimensional subspace, and then is integrated to convex segmentation functional to restrict the segmenting contour to a manifold of training shapes during the segmentation process. In contrast to existing statistic approaches of shape priors (uniform distribution, Gaussian distribution, KDE), RSDE can provide a high accuracy estimator of a probability density function which employs a small percentage of the available data sample.
The remainder of this paper is organized as follows. In Section 2, the shape prior based level set segmentation and RSDE are described. RSDE based on shape probabilistic representation for variational segmentation is presented in Section 3. The implementation procedure is proposed in Section 4 to minimize the energy. Experimental results are provided in Section 5, and the conclusions are given in Section 6.
Section snippets
Shape prior-based level set segmentation
Shape prior based image segmentation, which incorporates the shape prior information into the segmenting process, makes the final result more robust, accurate and efficient. In general, most level set segmentation models can be cast as variational minimization problem as followswhere ϕ is the level set function, and γ > 0 determines the relative importance of the two energy terms. Ei is a data-driven term which aims at driving the segmenting curve to the object boundaries, and E
A compact low-dimensional representation
It is well-known that statistical shape models can be performed more reliably and more efficiently in low-dimensional representations. For this reason, principal components analysis (PCA) of shape probabilistic representation (SPR) is used to capture the variations of shapes while removing redundant information. Using PCA, we compute the eigenmodes of the shape set χ = {q1, q2, ⋯qN}. We use only a subspace of χ spanned by the first eigenmodes {ψ1, ψ2, ⋯ψn}. The value of n must be chosen large
Energy minimization
In practice, the edge indicator term ∫ Ω|∇qα(Tρ(x))|dx in (16) can be neglected, since the proposed model is directly optimized in the linear subspace spanned by the principal components. Then, we get the following energy functional:
As for the region descriptors, several descriptors [13], [15], [24], [25] may be considered. In this work, we choose the following as the region descriptors.where po(u) and pb(u
Tracking a walking person
In this section, we apply the proposed model to track a walking person. Here four different data sets we used. One is used as training data, and the rest three data sets are used for testing purpose. The training set consists of 151 shapes from a sequence (showing a different person walking at a different pace), which is publicly available [20]. The RSDE is applied to this training set, and reduces the number of the kernels to 26 (near 17 percent of the original sample). Given an approximate
Conclusion
In this paper, a reduced set density estimator model based on shape probabilistic representation is constructed for image segmentation. If the sample size is very large, RSDE is an efficient approach to model nonlinear distributions of training shapes by providing a kernel density estimator which employs a small subset of the available data sample to provide similar levels of performance. In contrast to the commonly used signed distance functions, shape probabilistic representation can capture
Acknowledgments
This work was supported by a National Key Basic Research Project of China (973 Program No. 2012CB316400) and NSFC (No. 60872069).
References (36)
- et al.
Fronts propagation with curvature dependent speed: algorithms based on Hamilton–Jacobi formulations
Journal of Computational Physics
(1988) - et al.
Shape statistics in kernel space for variational image segmentation
Pattern Recognition
(2003) - et al.
Level Set Methods and Dynamic Implicit Surfaces
(2002) - et al.
Curve evolution implementation of the Mumford–Shah functional for image segmentation, denoising, interpolation, and magnification
IEEE Trans. Image Processing
(2001) - et al.
Statistical shape influence in geodesic active contours
IEEE International Conference on Computer Vision and Pattern Recognition
(2000) - et al.
A framework for image segmentation using shape models and kernel space shape priors
IEEE Transactions on Pattern Analysis and Machine Intelligence
(2008) - M. Rousson, N. Paragios, Shape priors for level set representations, in: Proceedings of the European Conference on...
- et al.
A shape-based approach to the segmentation of medical imagery using level sets
IEEE Transactions on Medical Imaging
(2003) - et al.
Towards recognition-based variational segmentation using shape priors and dynamic labeling
International Conference on Scale Space Methods in Computer Vision
(2003) - et al.
A multiphase dynamic labeling model for variational recognition-driven image segmentation”
International Journal of Computer Vision
(2006)
Level set based shape prior segmentation
IEEE International Conference Computer Vision and Pattern Recognition
Nonlinear dynamical shape priors for level set segmentation
Journal of Scientific Computing
Kernel density estimation and intrinsic alignment for shape priors in level set segmentation
International Journal of Computer Vision
Efficient kernel density estimation of shape and intensity priors for level set segmentation
International Conference on Medical Image Computing and Computer-Assisted Intervention
Geodesic active contours
International Journal of Computer Vision
Active contours without edges
IEEE Transactions on Image Processing
Implicit active shape models for 3d segmentation in MRI imaging
MICCAI
Probability density estimation from optimally condensed data samples
IEEE Transactions on Pattern Analysis and Machine Intelligence
Cited by (7)
Mammogram image visual enhancement, mass segmentation and classification
2015, Applied Soft Computing JournalCitation Excerpt :Variant feature transformation methods are mostly effective in segmenting micro-calcification along with mammogram masses and have been applied in [8,59,26,54,6,62,36] [57,44,55,1]. Other image segmentation methods that deal with color images include the work of [17,63,13]. Various mammogram image enhancement and segmentation techniques were proposed in literature such as the state-of-the-art approaches that include the work in [2,32,24,31].
Ultrasound kidney segmentation with a global prior shape
2013, Journal of Visual Communication and Image RepresentationCitation Excerpt :Afterwards, a prior shape is incorporated in a GAC model [16,17]. More related works can be found in [18–21]. In this paper, we replace the Rayleigh noise statistics used for prostate ultrasound images (the application in [35]) with the Fisher–Tippett statistics which are accepted to be appropriate for kidney ultrasound.
TRUS image segmentation with non-parametric kernel density estimation shape prior
2013, Biomedical Signal Processing and ControlCitation Excerpt :However, most of the methods are based on the assumption that the training shapes form a Gaussian distribution, while non-Gaussian shape prior used in several medical image segmentation tasks demonstrates superior results [11,12]. On the other hand, many studies have been carried out to embed the shape prior into both the explicit active contour model (ACM) [13] and the implicit level sets models [14], using non-Gaussian shape prior [11,15–25]. However, very few of them are specially designed for ultrasound images, and the prostate segmentation procedure cannot favor from the advantages of these methods for the following reasons.
Bayesian Robust Principal Component Analysis with Adaptive Singular Value Penalty
2020, Circuits, Systems, and Signal ProcessingImplicit kernel sparse shape representation: A sparse-neighbors-based objection segmentation framework
2017, Journal of the Optical Society of America A: Optics and Image Science, and VisionPCB CT image segmentation based on level set with shape prior
2016, Jisuanji Fuzhu Sheji Yu Tuxingxue Xuebao/Journal of Computer-Aided Design and Computer Graphics