HOW TO APPLY A MEAN FILTER FOR 3x3

I NEED TO APPLY THE 3x3 MEAN FILTER TO THE GRAY SCALE IMAGE. PLS SEND ME THE CODE FOR THAT

[Xheni]

thank you

Answer 1

[Wayne King]

How about just:

    h = 1/3*ones(3,1);
    H = h*h';
    % im be your image
    imfilt = filter2(H,im);

[ajith]

GREAT SIR THANKS A LOT

[suadad najim]

Can you tell me about the meaning of (ones) in your code please?

[Walter Roberson]

@suadad najim

    h = 1/3*ones(3,1);
    H = h*h';
    H

H = 3×3

0.1111 0.1111 0.1111 0.1111 0.1111 0.1111 0.1111 0.1111 0.1111

Notice this is the same as

    ones(3,3)/9

ans = 3×3

0.1111 0.1111 0.1111 0.1111 0.1111 0.1111 0.1111 0.1111 0.1111

which in turn is the same as

    [1/9, 1/9, 1/9;
     1/9, 1/9, 1/9;
     1/9, 1/9, 1/9]

ans = 3×3

0.1111 0.1111 0.1111 0.1111 0.1111 0.1111 0.1111 0.1111 0.1111

The ones() that @Wayne King used is a shorter way of writing out the individual components of the matrix.

See Image Analyst's discussion below https://www.mathworks.com/matlabcentral/answers/49226-how-to-apply-a-mean-filter-for-3x3#comment_2060839 about why you need to divide by 9 in this context.

[suadad najim]

@Walter Roberson Thank you for quick answer

I want you to see the image before and after applying the mean filter you gave. Is that normal?

Orginal image Resulted image

[Walter Roberson]

Check class() of the input image and class() of the output image. I suspect that class of the output is double and that you would get what you want if you uint8() that.

[DGM]

What is the class and scale of the input and output? The rendering of an image depends on an assumption of where white and black are. Floating-point images are expected to be unit-scale (i.e. 0-1). If you're using the above code, then it will be uint8-scale (0-255)

% this is a uint8 image
im = imread('cameraman.tif');

% create filter
h = 1/3*ones(3,1);
H = h*h';

% the output of this will be class 'double', but scaled as if uint8
imfilt = filter2(H,im);

% cast it to a class appropriate for its scale
imfilt = uint8(imfilt);

% display it
imshow(imfilt)

See also:

https://www.mathworks.com/help/images/convert-image-data-between-classes.html

Answer 2

[Image Analyst]

filteredImage = conv2(single(grayscaleImage), ones(3)/9);

[Sehairi K.]

filteredImage = conv2(grayscaleImage, ones(3)/9,'same');

[Image Analyst]

This will work if your grayscaleImage is already converted to double or single. Also, there are 3 edge options: 'same', 'full' (the default if none is supplied), and 'valid'. Usually for image processing the 'same' option is used.

[Atalay Asa]

How can we use it for 7x7 arithmetic mean filter?

[Image Analyst]

windowWidth = 7;
kernel = ones(windowWidth) / windowWidth^2;
outputImage = imfilter(grayImage, kernel);

[SP]

Hello, Sir. Can you explain to me that why the value in neighborhood need to be 1/9 (or 0.111)?

[Ridho Liwardana]

It's depend of matrix that you wanna use as filter, if it's a 3x3 matrix, the value is 1/(3x3) -> 1/9.

[Image Analyst]

@SP The reason you divide by windowWidth^2 (1/9 in this case) is because if you don't, the filtering is the sum of the filter window times the image to be filtered. So if the image were all 255 under the filter, you'd get an output value of 9*255 instead of 255. So the image would be 9 times as bright. To ensure that the filtered image lies in the same intensity range as the original image, you need to divide the filter values by windowWidth^2.

If you didn't have a uniform filter (like here where they are all 1's) you'd have to divide by the sum of the kernel:

% Divide unnormalized kernel by the sum of the values in it.
% This will ensure the output image has the same intensity range as the original.
kernel = kernel / sum(kernel(:)); 
outputImage = imfilter(grayImage, kernel);

Actually, that is the most general approach which will work for any filter.

Answer 3

[Mohamed Batran]

thank you for your reply simple but helped a lot for customizing the filter i want

Answer 4

[Celal Can Elgün]

How can we make it without using filter2(or any command) ?

[Walter Roberson]

http://uk.mathworks.com/matlabcentral/answers/38787-what-can-be-programmed-without-any-built-in-functions

Answer 5

[Geetha raja]

I NEED TO APPLY THE bilateral FILTER TO THE GRAY SCALE IMAGE for denoising. PLS SEND ME THE CODE FOR THAT

[Image Analyst]

Geetha, there is a part of the web site called "File Exchange". You can look there for community-contributed programs: Search for bilateral filter

Answer 6

[Jahid Hasan]

Sorry to ask here, how to write mean and median filter without using a in-built function? What are the way to do it? Any help or code syntax. Thank you

[Image Analyst]

You'd use a for loop in the obvious way where you just sum the elements and divide by the number of elements.

For median you'd need to write your own sorting routine (see Wikipedia or somewhere) and then take the middle element.

What can be programmed "without any built-in functions" ? - MATLAB Answers - MATLAB Central

[Jahid Hasan]

Thanks for your reply. Do you have any examples of it as a reference to understand it.

[Walter Roberson]

Potential reference implementation. Not tested.

sum_of_elements = 0;
index_of_element = 0;
while true
    index_of_element = index_of_element + 1;
    try
        sum_of_elements = sum_of_elements + YourVector(index_of_element);
    catch ME
        break;   %we ran off the end of the array
    end
end
number_of_elements = index_of_element - 1;  %loop over-counts by 1

mean_of_YourVector = sum_of_elements ./ number_of_elements;

sorted_vector = bogosort(YourVector, number_of_elements);

index_of_element = 0;
while true
    index_of_element = index_of_element + 1;
    if index_of_element + index_of_element == number_of_elements
        median_of_YourVector = YourVector(index_of_element);
        break
    elseif index_of_element + index_of_element + 1 == number_of_elements
        median_of_YourVector = (YourVector(index_of_element) + YourVector(index_of_element+1)) / 2;
        break
    end
end

function sortedVector = bogosort(vectorToSort, number_of_elements)
  %this is a legitimate sort
  while true
      neworder = randperm(number_of_elements);
      sortedVector = vectorToSort(neworder);
      
      index_of_element = 0;
      while true
          index_of_element = index_of_element + 1;
          if index_of_element >= number_of_elements
              %we got to the end of the list without detecting out-of-order
              return;
          end
          if YourVector(index_of_element) > YourVector(index_of_element+1)
              %detected something out of order, try again with a new order
              break
          end
      end
  end
end

[DGM]

As far as sliding filter/convolution examples go, here are some links. I'm sure there are others. These are just the ones I have in my notes.

general sliding/moving window filter base

https://www.mathworks.com/matlabcentral/answers/1466576-changing-a-3x3-to-5x5-median-filter

same thing only with a gaussian mean setup

https://www.mathworks.com/matlabcentral/answers/1648495-i-am-trying-to-write-a-program-for-a-grayscale-image-blurring-without-using-gaussian-or-conv2-functi

basic mean filter example

https://www.mathworks.com/matlabcentral/answers/1571763-rgb2gray-by-5-5-vicinity-marix

https://www.mathworks.com/matlabcentral/answers/1632145-mean-filter-with-kernel-size

basic 2D convolution build (full size)

https://www.mathworks.com/matlabcentral/answers/820635-how-to-convolve-two-matrices-with-different-sizes

basic 2D convolution build (same size)

https://www.mathworks.com/matlabcentral/answers/791809-2d-convolution-with-same-dimensions