Speed up convolution and supersampling in 3D binary matrix

HAllo! and thanks in advance to averybody.

I have a 3D matrix made by confocal microscopy images that have the same resolution on x,y axes and about 1/7 of resolution on z axes. To perform a supersampling on z axes I wrote this code that perform a convolution with a gaussian function on Z axes and than an uppersampling for each Z vector.

    function [matrixZ,Zplanes] = denoising3Z (matrix,planes); 
    x = [1,1,1];
    h = @(x) gaussmf(x,[4,0]);
    Zplanes = planes * 7;
    [r,c] = size(matrix{1,1});
    matrixZ = zeros (r,c, Zplanes);
    for i = 1: r
        for j = 1 : c
            vectorZ = zeros(1,6);
            for z = 1 : planes
            vectorZ(1,z) = matrix{1,z}(i,j);
            end
            vectorCZ(1,:) = filter(h(x),2,vectorZ(1,:));
            V2 = imresize(vectorCZ, [1 Zplanes], 'bilinear');
            matrixZ(i,j,:) = V2(1,:);
        end
       i     
    end
    end

It works very well but it has to repeat these operation for 6 million of Z vector per images and the entire operation takes a long time. How I can speed up the process? It is possible to create a filter volume and applicate the volume to the matrix in just one pass? Probably it should be a bit more rapid. Thanks.

Answer 1

[Emanuele Gandola]

Thank you Deepak, I found this really usefull solution

 function [matrixZ,Zplanes] = denoising3D (matrix,planes); 

 Zplanes = planes * 7;

 B = imgaussfilt3(matrix);

 [r,c,z] = size(B);

 matrixZ = zeros (r,c, Zplanes);

 for i = 1: r
         img(:,:) = B(i,:,:);   
        imgZ = imresize(img, [c Zplanes]);
        matrixZ(i,:,:) = imgZ;

 end
 end

imgaussfilt3 is a new volumetric function implemented in the last version of image processing toolbox that is brilliant! And for the rescalig I threated the volume as a group of slice using imresize. In this way I reduced the computational time of 100 about times

Answer 2

[Deepak Bhatia]

Looking at the nested loop structure, I recommend looking into vectorization in MATLAB as it might reduce code execution time.