A fuzzy variational model for segmentation of images having intensity inhomogeneity and slight texture
- PDF / 8,561,534 Bytes
- 16 Pages / 595.276 x 790.866 pts Page_size
- 39 Downloads / 148 Views
METHODOLOGIES AND APPLICATION
A fuzzy variational model for segmentation of images having intensity inhomogeneity and slight texture Ali Ahmad1 · Noor Badshah1 · Haider Ali2
© Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Segmentation of real images having unwanted outliers, inhomogeneity or complex background is always very challenging for active contour models. In this paper, we propose a novel model for segmentation of such type of images. The proposed model is based on fuzzy energy functional, which uses coefficient of variation as a region statistics. The proposed model is convex due to introduction of fuzzy membership functions in the energy functional and hence converges to the absolute minima and avoids local minima. Convexity of the proposed model is proved, and hence, the model is independent of initial placement of the contour. Experimental results of the proposed model are compared with other state-of-the-art existing models both qualitatively and quantitatively. For quantitative comparison, we have used Jaccard similarity index and computational complexity. The proposed model is tested on various data sets containing noisy images, images having intensity inhomogeneity and slight texture. In all experimental results, performance of the proposed model can be seen in the experimental section. Keywords Image segmentation · Fuzzy sets · Pseudo-level set · Kernel metric
1 Introduction In recent years, active contour methods have got much attention, and have found wide range of applications in problems including image segmentation and visual tracking (Paragios et al. 2005; Paragios and Deriche 2000; Zhang and Freedman 2003). The main idea behind active contour methods is to allow a contour to evolve in order to obtain the desired segmentation (Morel and Solimini 2012; Sethian 1999). Active contour models can be classified into two classes, say edgebased and region-based models. Edge-based active contour models usually use an edge detector function depending on the image gradient to stop the evolving curve on the object boundary (Caselles et al. 1993; Kass et al. 1988; Malladi et al. 1993). These models have few weaknesses like sensitivity to noise, initialization of the active contour and may fail to segment images havCommunicated by V. Loia.
B
Noor Badshah [email protected]
1
University of Engineering and Technology Peshawar, Peshawar, Pakistan
2
University of Peshawar, Peshawar, Pakistan
ing objects with blur or weak boundaries. On the other hand, region-based models use region information (region statistics) for evolution of contour to segment images (Caselles et al. 1997; Yezzi et al. 1997). One of the most known regionbased variational model is the Mumford and Shah model (1989), which finds an optimal piecewise smooth function to approximate the observed image. This model is theoretically strong model, but it is hard to implement it computationally in practice; therefore, a first variation of the MS model was proposed by Chan and Vese (CV) (2001); a level-set metho
Data Loading...
