how to make a filter to the noisy image? so that the noise will be removed

filter

Answer 1

[Walter Roberson]

How do you know that it is noisy? How do you know that the image should not be exactly like it looks? Do you have another image to compare against?

Astronomers used to curse at the smudges in the high-depth-of-field images, curse at the "noisy" images that were preventing them from getting nice crisp backgrounds. Then someone started doing a study of the "noise" and discovered that it was the Cosmic Background Radiation and in fact had a lot of information in it about the early universe. Now they send up satellites designed to filter out the stars so they can get good images of what used to be considered noise.

Answer 2

[Image Analyst]

There may be more ways to remove noise in images than stars in the universe. People have been working on it for ages and come up with dozens of papers every year on it. See section 5.3 here in Vision Bib.

You can do linear filtering with conv2() such as an averaging or box filter. Another very simple filter (non-linear) is the median filter, accomplished by medfilt2() in the Image Processing Toolbox. If those are no good then you move on to more sophisticated non-linear filters like bilateral, k nearest neighbors, mean shift, etc. And there are even more sophisticated methods and complicated methods like BM3D (last I heard BM3D was the current champion), non-local means, K-SVD, K-LLD, UINTA, etc.