# Asymmetric 3D Gaussian Filtering in Matlab

10 views (last 30 days)
saipb on 2 Aug 2019
Commented: saipb on 6 Aug 2019
Hello,
I have a flowfield that has u,v,w velocities. Each of them are 3D arrays - say 480 x 480 x 60. I know that we can do an asymmetric Gaussian filtering in 3D in the following manner:
sigma = [10, 5, 3];
u_filt = imgaussfilt3(u,sigma);
The problem that I am facing is that my grid is non-unform. That is while my dx = dy = 1km, my dz changes as a function of height. How do I modify the imgaussfilt3 such that the sigma_z is a function of z? One way is to call the imgaussfilt 60 times (that is the size of z) but that would be very inefficient. Is there an alternative solution?
Sincerely,
Sai

Navya Seelam on 6 Aug 2019
Hi,
The input argument “sigma” to the function “imgaussfilt3” must be a positive number or a 3-element vector of positive numbers. In case if the sigma_z is a function of z, perform the filtering iteratively calling the “imgaussfilt3” 60 times.

#### 1 Comment

saipb on 6 Aug 2019
Thank you for answering this, Navya. At present, I am calling imgaussfilt3 60 times and each time, I extract the filtered output at that particular 'z'. In the end, the new filtered flow field is the synthesis of the 60 filters done level by level. Is this the best way in terms of computational time? Is there a way to write a user-defined 3D spatial, Gaussian filter that is asymmetric in 'z' that is more computationally efficient than my current solution?