Finding the average of the neighbors of each pixel in an image

I want to replace each pixel by the average of 3 x 3 neighbors using 4-connected neighbors (I4) and 8- connected neighbors (I8).

Answer 1

[Matt J]

mean4=conv2(image,[0 1 0; 1 1 1; 0 1 0]/5,'same');
mean8=conv2(image,ones(3)/9,'same');

[Srinand Andey]

Hey can you just explain me what actually happens in those lines!

[Jan]

Did you read the documentation of conv2 already?

doc conv2

[Srinand Andey]

my problem is i have an image matrix with size 426x422x3 uint8 and i have to replace each pixel by the average of 3 x 3 neighbors using 4-connected neighbors (I4). And yeah i read te documentation but i wasn't able to solve this part.

[Jan]

What exactly is not clear yet? The convolution is multiplying the submatrices with the given matrix and calculating the sum. [0 1 0; 1 1 1; 0 1 0] adds the 4-connected neighbors to the center and dividing by 5 is the normalization.

[Image Analyst]

@Srinand Andey, so you have a volumetric image. Each voxel has 26 8-connected neighbors (each voxel in a 3x3x3 cube except for the center voxel), and 6 4-connected neighbors (each face of a 3x3x3 cube). Do you want the average of the 6 nearest neighbors? Because we're not sure what you mean by 3x3 (instead of 3x3x3) when you have a 3-D image.

[Srinand Andey]

this is the image i used and when i check the size in the matlab it shows

when i run the code

[Image Analyst]

That's a color image. You need to turn it into a gray scale image. Fastest way is to just take the red channel:

img = img(:, :, 1); % Now img will be a gray scale image that is the red channel of the old color image.

Also, you need to make sure the image is double before you pass it into conv2(). You cannot use a uint8 image or you'll get an error.

[Matt J]

Use convn() instead.

[Matt J]

You cannot use a uint8 image or you'll get an error.

That may depend on the Matlab version:

 image=imread('peppers.png');
 mean4=convn(image,[0 1 0; 1 1 1; 0 1 0]/5,'same');
 mean8=convn(image,ones(3)/9,'same');

whos

  Name         Size                 Bytes  Class     Attributes

  image      384x512x3             589824  uint8               
  mean4      384x512x3            4718592  double              
  mean8      384x512x3            4718592  double              

[Image Analyst]

Matt, you're right. They seem to have "fixed" that at some point.

[Srinand Andey]

i am able to see the values getting changed to mean but when i try to display the image after the mean using imshow() a blank image is being displayed

[Image Analyst]

It's a floating point image so you need to use [] because the range is beyond the 0-1 that imshow() expects for a floating point image.

mean4=conv2(yourImage,[0 1 0; 1 1 1; 0 1 0]/5,'same'); % Don't use image as a variable name because it's a built-in function name.
imshow(mean4, []);

[Isabella]

I want to replace each pixel by the average of 3 x 3 neighbors using 4-connected neighbors (I4) and 8- connected neighbors (I8).

What is the advantage or strategy behind using I4 or I8? I want to replace each pixel by the avg of its 3x3 neighbors, I don't mind how i do it. can you advise?

[Image Analyst]

@Mercedes you must know if you want to do it, right?

[Isabella]

@Image Analyst Can I replace each pixel by the avg of its 3x3 neighbors, without I4 or I8?

[Image Analyst]

You have to specify which pixels to use in the average. That kernel basically specifies which of the 9 pixels in the 3x3 window to use in computing the average. You can specify whatever you want in your kernel. Like if you wanted to replace the center pixel by the average of the 4 on the corners, you'd do

kernel = [1,0,1; 0,0,0; 1,0,1] / 4; % Diagonals only.
blurredImage = conv2(double(grayImage), kernel, 'same');
% or, including the middle pixel:
kernel = [1,0,1; 0,1,0; 1,0,1] / 4; % Diagonals plus center pixel.
blurredImage = conv2(double(grayImage), kernel, 'same');