Adaptive variational sinogram interpolation of sparsely sampled CT data

We present various kinds of variational PDE based methods to interpolate missing sinogram data for tomographic image reconstruction. Using the observed sinogram data we inpaint the projection data by diffusion. To overcome the problem of contour blurring we consider nonlinear and anisotropic diffusion based regularizes and include optical flow information in order to preserve the sinusoidal traces corresponding to object contours in the reconstructed image. We compare our results to a spectral deconvolution based interpolation and show that the method can easily be extended to 3D