Gabor feature space diffusion via the minimal weighted area method

Abstract

Gabor feature space is elaborated for representation, processing and segmentation of textured images. As a first step of preprocessing of images represented in this space, we introduce an algorithm for Gabor feature space denoising. It is a geometric-based algorithm that applies diffusion-like equation derived from a minimal weighted area functional, introduced previously and applied in the context of stereo reconstruction models [6,12]. In a previous publication we have already demonstrated how to generalize the intensity-based geodesic active contours model to the Gabor spatial-feature space. This space is represented, via the Beltrami framework, as a 2D Riemannian manifold embedded in a 6D space. In this study we apply the minimal weighted area method to smooth the Gabor space features prior to the application of the geodesic active contour mechanism. We show that this ”Weighted Beltrami” approach preserves edges better than the original Beltrami diffusion. Experimental results of this feature space denoisingpro cess and of the geodesic active contour mechanism applied to the denoised feature space are presented.