3D Data Denoising and Inpainting with low redundancy 3D Fast Curvelets


2/ Experiments

Here are the results of some experiments carried out to demonstrate the efficiency of such transform used in iterative thresholding algorithms to deal with inverse problems.

a) Fig 4 shows the inpainting of a synthetic cerebra MRI from BrainWeb (http://www.bic.mni.mcgill.ca/brainweb/), with 80% random missing voxels or 10% random missing slices.

Fig. 4 - Inpainting of MRI data

b) Video sequence inpainting with the LR-FastCurvelets and wavelets, the mask being a rotating swirl, and a few missing frames

Fig. 5a - Original central frame

Fig. 5d - Original missing frame

Fig. 5e - Recovered missing frame

Fig. 5b - Masked data

Fig. 5c - Recovered central frame