Smoothing Tensor-Valued Images Using Anisotropic Geodesic Diffusion

Fan Zhang, Edwin R. Hancock

Abstract: This paper considers the feature space of DT-MRI as a differential manifold with an affine-invariant metric. We generalise Di Zenzo’s structure tensor to tensor-valued images for edge detection. To improve the quality of the edges, we develop a generalised Perona-Malik method for smoothing tensor images. We demonstrate our algorithm on both synthetic and real DT-MRI data.