How can i perform histogram equalization transformation in 3 by 3 neighborhood of a pixel and repeat the process?

Say for example

X=[2 1 0; 0 1 3; 4 1 4];

%zero padding

A=padarray(X,[1,1]);

Now for each pixel of X a 3 by 3 neighborhood matrix can be generated from A.

e.g.

A(1:3,1:3);  A(1:3,2:4);

For each window, i want to use histogram equalization transformation for the central pixel and repeat the process up to

length(x(:)).

The steps of my algorithm are summarized here

1. Find the central pixel and its neighborhood.
2. Use histogram equalization transformation for each neighborhood to transform the central pixel.
3. Move from pixel to pixel and repeat the process.

Answer 1

[Image Analyst]

You need to use adapthisteq(). It will do that, though for such a small window your output image is likely to look list noise.

[Tuhin Sheikh]

Dear sir,

it was a hypothetical example to understand the algorithm.

adapthisteq() perform local histogram equalization along with it limits the contrast to overcome over-enhancement.

Actually i'm planning to modify local histogram equalization algorithm to local histogram matching with any specific distribution.

Thanks for the suggestion for using adapthisteq()

Could you please help me further, how can i match with normal distribution(with specification of parameter) other than 'uniform','rayleigh', 'exponential'?

Thanks

[Image Analyst]

You can't match distributions with only 9 samples - there is not statistically enough samples to tell the difference between the various distributions you list. Anyway, do you think it would make any visible difference? I kind of doubt it.

[Tuhin Sheikh]

Dear sir,

i was trying the program with small matrix first.

Definitely what you said is true.

say this is my image and i would apply the transformation function on 8 by 8 matrix

x=imread('tire.tif');
adapthisteq(x) % it has distributional specification for Rayleigh, exponential and uniform

my question was is it possible to match with normal distribution in this function?

[Image Analyst]

It doesn't look like it. Why do you want it? What do you think the resulting image will look like and why do you expect it would be preferable to those distributions? If you really need it for some reason, you may have to program it up yourself.