Non Local Spatial Regularization NNLS

Hi everyone,

My problem is the following one : I'm trying to build a non local spatial regularization for NNLS algorithm, and I have difficulties in implementing it. In fact, I try to adapt a method found in the litterature (script : http://www.mathworks.com/matlabcentral/fileexchange/13176-non-local-means-filter/content/NLmeansfilter.m), except that I don't have a scalar mapping (in regards to 2D images), but a discrete distribution inside each pixels. As a consequence, and I wont go any further into details, I would like to select one squared neighboorhood of pixels (7x7) in my image, and compares it to other squared neighboorhood nearby inside a 21x21 search window. The result (whatever application I have to compute - this one is well implemented) remain in the central voxel of the 1st square. My adaptation of the previous script is extremly slow (around 2h30 for a 192x192 image)

I looked on the proposed implementation by G. PEYRE (https://www.ceremade.dauphine.fr/~peyre/numerical-tour/tours/denoisingadv_6_nl_means/ - really tricky work, it's wonderful), which allow one to rapidly select such square with a simple coordinates (i,j) on the image, but I can't reproduce it with a third dimension (my discrete distribution).

Does anyone have any idea about this ? Regards