MATLAB Answers

0

how to smooth data?

Asked by Dimitrios on 28 Oct 2014
Latest activity Answered by David Young
on 28 Oct 2014
From a simulation i got some output where there are some spikes that not correspond with the other data.Which is the best way to smooth them.Both in 2d and 1d matrix.I provide some examples to be more clear:
This is for 2d data and a following with a 1d data:
And i want to find a way without knowing where exactly are the spices.Cound be in different index depending on the case.Any Idea?Are there some kind of methods? Thanks in advance

  0 Comments

Sign in to comment.

1 Answer

David Young
Answer by David Young
on 28 Oct 2014

It's impossible to say what the best method is without a model of spike formation and a function that quantifies the cost of residual errors in the smoothed data. Without these, you could try some techniques and try to assess the performance.
There are many different possibilities. Every method that reduces the noise will do some damage to the data, and the problem is to find which is the best trade-off.
A reasonable starting point might be median filtering, using medfilt2() and medfilt1(). You could also look at more general nonlinear filters, which you can implement using nlfilter().
Another possibility is linear filtering. For example, you can smooth with a Gaussian kernel using fspecial() and conv2(). There's an easy interface in Gradients with Gaussian Smoothing in the File Exchange.
There are many more advanced methods if these aren't adequate, but it will involve experiment and research to get it right.

  0 Comments

Sign in to comment.