This is likely due to the adaptive HOSVD bases derived from the stacked patches, which are more suitable for reconstruction.
The proposed HOSVD methods outperformed the nonlocal upsampling method in most of the diffusion directions and the HOSVD-M method achieved the best results in most of them.
The proposed method introduced HOSVD into the SR framework as a regularization term to achieve better image reconstruction and more efficient computation.
Compared with conventional interpolation and patch-based SR methods, the improvements made by the proposed HOSVD method can be contributed to two features.
This increase in speed is likely due to an inherent dimensional decreasing property of the SVD as well as HOSVD.
The adaptive HOSVD bases acquired from the image ensured a more accurate image reconstruction and manipulation of similar patch stacks led to a reduction in computational complexity.
a) The original dataset, the phantom datasets reconstructed using (b) B-splines, (c) the nonlocal method, (d) the proposed HOSVD, and (e) the proposed HOSVD-M.
Results of (c) B-spline reconstruction, (d) nonlocal method, (e) the proposed HOSVD, and (f) the proposed HOSVD-M.
a) FA maps estimated using the gold standard; FA maps obtained for the reconstructed dataset using the (b) B-spline, (c) nonlocal method, (d) the proposed HOSVD, and (e) the proposed HOSVDM.
Caption: Figure 6: (a) FA colormap for the gold standard; FA colormaps for the reconstructed dataset using (b) B- spline, (c) nonlocal method, (d) the proposed HOSVD, and (e) the proposed HOSVD-M.