MR Image Segmentation Using Active Contour Model Incorporated with Sobel Edge Detection
This paper proposes a segmentation method which combines Active contour model with Sobel edge detection. The introduction of distance regular-ized formulation eliminates the need for reinitialization when we minimize the energy function by using the level
- PDF / 1,164,509 Bytes
- 9 Pages / 439.37 x 666.14 pts Page_size
- 99 Downloads / 191 Views
Abstract. This paper proposes a segmentation method which combines Active contour model with Sobel edge detection. The introduction of distance regularized formulation eliminates the need for reinitialization when we minimize the energy function by using the level set method. We test our method on MR image and compare it with several methods in the literature. The results achieved are better than the ones of existing techniques, showing the effectiveness of the proposed method. Keywords: Image segmentation · Active contour model · Edge detection operator · Level set method
1
Introduction
Image segmentation has been widely used in medical analysis such as identifying tumors or soft tissue injuries from medical illustration [1], [2]. Active contour model is an excellent method and has been applied to image segmentation [3], [4], [11]. The general idea is to first initialize a curve around the object and then makes the curve move toward the object’s interior morphology under the control of an energy function and finally stop at the boundary of the target area. Compare with classical image segmentation methods, active contour model has two advantages. Firstly, active contour model can be easily derived by energy minimization framework [5], [6]. Secondly, active contour model is able to achieve the sub-pixel accuracy of object boundaries [7]. It can be divided into two types: parametric active contour model and geometric active contour model. Parametric active contour was introduced by Kass et. al [8]. The energy function is minimized to attract the contour toward the edges, in which first derivative and second derivative were used to control the smoothness of the active contour, and therefore their method belongs to parametric active contour. However, the main disadvantage of the parametric active contour models is that the relation between the parametrization of the contour and geometry of the objects is not obvious. To overcome the drawback of the parametric active contour model, geometric active contour model was proposed, which can be further categorized into edge-based models and region-based models. Edge-based models use edge information to attract © Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part I, CCIS 546, pp. 429–437, 2015. DOI: 10.1007/978-3-662-48558-3_43
430
H. Zhang et al.
the active contour move toward the object boundaries. But it still remains a challenge to find a proper trade-off between noise smoothing and edge information preservation, especially in the real condition. While, most MR image is noisy, if the isotropic smoothing such as Gaussian is strong, the edge would be smooth too. Region-based models used a certain region descriptor to guide the motion of the active contour, and therefore it could detect contours without edges. However, the region-based models are limited by intensity homogeneity. In fact, intensity inhomogeneity often occurs in MR images. Tsai et al. [9] proposed region-based model which regards image segmentation as a problem of finding a best
Data Loading...
