Segmenting Multiple Textured Objects Using Geodesic Active Contour and DWT
We address the issue of segmenting multiple textured objects in presence of a background texture. The proposed technique is based on Geodesic Active Contour (GAC) and can segment multiple textured objects from the textured background. For an input texture
- PDF / 6,074,903 Bytes
- 8 Pages / 430 x 660 pts Page_size
- 99 Downloads / 193 Views
Abstract. We address the issue of segmenting multiple textured objects in presence of a background texture. The proposed technique is based on Geodesic Active Contour (GAC) and can segment multiple textured objects from the textured background. For an input texture image, a texture feature space is created using scalogram obtained from discrete wavelet transform (DWT). Then, a 2-D Riemannian manifold of local features is extracted via the Beltrami framework. The metric of this surface provides a good indicator of texture changes, and therefore, is used in GAC algorithm for texture segmentation. Our main contribution in this work lie in the development of new DWT and scalogram based texture features which have a strong discriminating power to define a good texture edge metric which is used in GAC technique. We validate our technique using a set of synthetic and natural texture images. Keywords: Snake, segmentation, texture, DWT, scalogram.
1
Introduction
Active contours are extensively used in the field of computer vision and image processing. In this paper, we present a texture object segmentation technique which is based on Geodesic Active Contour (GAC) [1] and discrete wavelet transform (DWT) based texture features, and can segment multiple textured objects from the textured background. Our algorithm is based on the generalization of the GAC model from 1-D intensity based feature space to multi-dimensional feature space [2]. In our approach, image is represented in a n-dimensional texture feature space which is derived from the image using scalograms [3] of the DWT. We derive edge indication function (stopping function) used in GAC from the texture feature space of the image, by viewing texture feature space as Riemannian manifold. Sochen et al. [4] showed that the images or image feature spaces can be described as Riemannian manifolds embedded in a higherdimensional space, via the Beltrami framework. Their approach is based on the polyakov action functional which weights the mapping between the image manifold (and its metric) and the image features manifold (and its corresponding metric). In our approach, a 2-D Riemannian manifold of local features is extracted from the texture features via the Beltrami framework [4]. The metric of this surface provides a good indicator of texture changes, and therefore, is used A. Ghosh, R.K. De, and S.K. Pal (Eds.): PReMI 2007, LNCS 4815, pp. 111–118, 2007. c Springer-Verlag Berlin Heidelberg 2007
112
S. Prakash and S. Das
in GAC for texture segmentation. The determinant of the metric of this manifold is interpreted as a measure of the presence of the gradient on the manifold. Similar approaches where the GAC scheme is applied to some feature space of the images, were studied in [5,6,7]. The aim of our study is to generalize the intensity based GAC model and apply it to DWT and scalogram based wavelet feature space of the images. Our main contribution in this work lie in the development of new texture features which give a strong texture discriminating power and in turn
Data Loading...
