A Computational Introduction to Digital Image Processing, 2nd Edition

Neighborhood Processing 115

p

B

D

A

C

FIGURE 5.22: Alternative neighborhoods for a Kuwahara filter

MATLAB/Octave

>> cd = float(c);

>> cdm = imfilter(cd,ones(3)/9,’symmetric’);

>> cd2f = imfilter(cd.^2,ones(3)/9,’symmetric’);

>> cdv = cd2f - cdm.^2;

Python

In: cd = float32(c)

In: cdm = ndi.uniform

_

filter(cd,(3,3))

In: cd2f = ndi.uniform

_

filter(cd

**

2,(3,3))

In: cdv = cd2f - cdm

**

2

At this stage, the array cdm contains all the mean values, and cdv contains all the variances.

At every point (i, j) in the image then:

1. Compute the list

vars = [cdv(i − 1, j − 1), cdv( i − 1, j + 1), cdv(i + 1, j − 1), cdv(i + 1, j + 1)] .

2. Also compute the list

means = [cdm(i − 1,