Fast 3D Mean Shift Filter for CT Images

In this paper we investigate the ability of the mean shift (MS) algorithm for denoising of 3D Computer Tomography (CT) data sets. The large size of the volume data sets makes it infeasible to apply a 3D version of the MS algorithm directly. Therefore, we introduce a variant of the MS algorithm using information propagation. We would like to make use of the 3D nature of the data with a considerably reduced running time of the algorithm. The proposed version is compared to a 2D implementation of the same algorithm applied slice by slice and other filter methods such as median filter and bilateral filtering. The advantages and disadvantages of each algorithm are shown on different CT data sets.