Diffusion tensor images edge-directed interpolation

It has been demonstrated that, for scalar images, edge-directed interpolation techniques are able to produce better results, both visually and quantitatively, than non-adaptive traditional interpolation methods. We have extended the edge-directed concept to the interpolation of multi-valued diffusion tensor images. The interpolation framework is based on the improved edge-directed scalar image interpolation, which does not require estimation of local edges, and which interpolates the tensors in their mathematically native Riemannian space in order to respect tensor geometry of positive semi-definiteness. The framework has been tested on phantom scalar images and real diffusion tensor images, and is shown to be able to achieve better results than the Euclidean non-adaptive methods.