Fig. 2From: Postreconstruction filtering of 3D PET images by using weighted higher-order singular value decompositionFlow diagram of the proposed HOSVD-KSW algorithm. Note that, the reference patches were obtained by certain voxels for saving execute time. A 4D stack was formed for every reference patch, by clustering itself and its similar patches, with a weighted manner. The stack was then processed orderly by HOSVD, truncated transformation coefficients, and inverse HOSVD, to obtain a estimation value of voxels included in the stackBack to article page