How to apply Gaussian blur to a specific area (oval) in an image?

Hi. I have an image processing problem. I have images of faces (JPG, freely available for research purposes) which I inserted in gray ovals (see example below). All images share the same dimensions. My question: how can you apply Gaussian blur to the pixels around the edge of the oval, i.e., where the pixels of the oval edge and the face image meet. My aim is to decrease contrast in the specific perimeter so the faces look embedded in their gray background.

Can someone help me with this?

This is the script I wrote to add the gray oval, but I haven't figured out how to add the Gaussian blur:

% Specify the input and output directories
input_dir = '/X';
output_dir = '/Y';
% Get a list of all the JPG files in the input directory
file_list = dir(fullfile(input_dir, '*.jpg'));
% Loop over each file in the list
for i = 1:numel(file_list)
    % Load the image
    img = imread(fullfile(input_dir, file_list(i).name));
    
    % Get the size of the image
    [rows, cols, ~] = size(img);
    
    % Create a meshgrid of x and y coordinates
    [x, y] = meshgrid(1:cols, 1:rows);
    
    % Calculate the x and y coordinates of the center of the image
    center_x = (cols+1)/2;
    center_y = (rows+1)/2;
    
    % Calculate the radius of the oval
    radius_x = center_x - 1;
    radius_y = center_y - 1;
    
    % Create a mask for the oval using the x and y coordinates, center, and radii
    mask = ((x-center_x)/radius_x).^2 + ((y-center_y)/radius_y).^2 <= 1;
    
    % Create a copy of the original image with the contour region masked out
    masked_img = img;
    masked_img(repmat(~mask, [1, 1, 3])) = 220;
    
    % Construct the output filename with 'REG' at the end
    [~, filename, ext] = fileparts(file_list(i).name);
    output_filename = [filename, 'REG', ext];
    
    % Save the masked image to the output directory with the modified filename
    imwrite(masked_img, fullfile(output_dir, output_filename));
end

Answer 1

[DGM]

ideally, you'd do this during composition instead of afterwards. Doing it afterwards is double the work.

Here is one example

https://www.mathworks.com/matlabcentral/answers/1448374-applying-a-circular-mask-to-an-image

And here's a more rudimentary example without specialized tools

https://www.mathworks.com/matlabcentral/answers/890417-i-want-to-remove-black-color-from-my-image-and-make-the-background-transparent-so-that-i-can-paste-i

The rest is the task of generating the mask.

Here is an example of doing the composition, starting with an image, a background color, and a logical mask.

% load foreground image
FG = imread('cameraman.tif'); % raw single-channel source

% background color
BG = [0.4 0.04 0.25]; % any color tuple

% mask parameters
center = [128 128]; % [x y]
radius = 128;
blurwidth = 30;

% i assume you start with a logical mask
% you can create this interactively with drawcircle() or drawellipse()
% but i'm going to do it programmatically
imshow(FG);
ROI = images.roi.Circle(gca);
ROI.Center = center;
ROI.Radius = radius-blurwidth/2; % create the mask undersize
mask = createMask(ROI);

% soften the mask
mask = imgaussfilt(double(mask),blurwidth/4);

% compose the foreground image with the solid-color background
outpict = im2double(FG).*mask + permute(BG,[1 3 2]).*(1-mask);
outpict = im2uint8(outpict);

imshow(outpict,'border','tight')

Note that I created the mask undersize to begin with instead of eroding it. This is simpler and helps avoid complications if the mask ever gets close to the image border.

EDIT: This is an example doing the same thing with an ellipse ROI object instead

% load foreground image
FG = imread('cameraman.tif'); % raw single-channel source

% background color
BG = [0.4 0.25 0.04]; % a tuple

% mask parameters
center = [128 128]; % [x y]
semiaxes = [150 90];
angle = -45;
blurwidth = 30;

% i assume you start with a logical mask
% you can create this interactively with drawcircle() or drawellipse()
% but i'm going to do it programmatically
imshow(FG);
ROI = images.roi.Ellipse(gca);
ROI.Center = center;
ROI.SemiAxes = semiaxes-blurwidth/2;
ROI.RotationAngle = mod(angle,360);
mask = createMask(ROI);

% soften the mask
mask = imgaussfilt(double(mask),blurwidth/4);

% compose the foreground image with the solid-color background
outpict = im2double(FG).*mask + permute(BG,[1 3 2]).*(1-mask);
outpict = im2uint8(outpict);

imshow(outpict,'border','tight')

[DGM]

Note that in these examples, I'm blurring the mask, not the image. The image content is not changed spatially, it just fades out.

Let's say you didn't want to recompose the image. You wanted to blur an annular region in the output image to hide the mask transition. This is what the question asks for and what others are suggesting. What would happen if you did that?

% load composed image
FG = imread('op1.png');

% assuming mask parameters are already known!
% mask parameters
center = [128 128]; % [x y]
semiaxes = [155 90];
angle = -45;
blurwidth = 30;

% i assume you start with a logical mask
% you can create this interactively with drawcircle() or drawellipse()
% but i'm going to do it programmatically
imshow(FG);
ROI = images.roi.Ellipse(gca);
ROI.Center = center;
ROI.SemiAxes = semiaxes-blurwidth/2;
ROI.RotationAngle = mod(angle,360);
mask = createMask(ROI);

% blur the image, not the mask
BG = imgaussfilt(im2double(FG),blurwidth/4);

% compose the foreground image with the blurred background
outpict = im2double(FG).*mask + BG.*(1-mask);
outpict = im2uint8(outpict);

imshow(outpict,'border','tight')

So long as your masking is logical, you will always have noticeable transitions. Note also that the blurring moved image content such that dark regions now intersect the mask edge and cause a distinct step.

You could blur both the image and the mask, but you have to blur at least the mask. If you did both, this is what you'd get.

Be aware though that doing both is not the same as a variable blur. See also:

https://www.mathworks.com/matlabcentral/answers/1651050-creating-a-blur-brush-tool-for-images#answer_951155

[Neuro]

Thanks! This is very helpful. This is the script I came up with. It draws the oval correctly, but I haven't figured out how to add the Gaussian filter in a way that is accurate. What would you recommend me to do?

% Specify the input and output directories
input_dir = '/X';
output_dir = '/Y';

% Get a list of all the JPG files in the input directory
file_list = dir(fullfile(input_dir, '*.jpg'));

% Loop over each file in the list
for i = 1:numel(file_list)
    % Load the image
    img = imread(fullfile(input_dir, file_list(i).name));
    
    % Get the size of the image
    [rows, cols, ~] = size(img);
    
    % Create a meshgrid of x and y coordinates
    [x, y] = meshgrid(1:cols, 1:rows);
    
    % Calculate the x and y coordinates of the center of the image
    center_x = (cols+1)/2;
    center_y = (rows+1)/2;
    
    % Calculate the radius of the oval
    radius_x = center_x - 1;
    radius_y = center_y - 1;
    
    % Create a mask for the oval using the x and y coordinates, center, and radii
    mask = ((x-center_x)/radius_x).^2 + ((y-center_y)/radius_y).^2 <= 1;
    
    % Create a copy of the original image with the contour region masked out
    masked_img = img;
    masked_img(repmat(~mask, [1, 1, 3])) = 220;
    
    % Construct the output filename with 'REG' at the end
    [~, filename, ext] = fileparts(file_list(i).name);
    output_filename = [filename, 'REG', ext];
    
    % Save the masked image to the output directory with the modified filename
    imwrite(masked_img, fullfile(output_dir, output_filename));
end

[DGM]

Try this. I removed some parts so that I can run it here as a demo.

% parameters
blurwidth = 20;
bgcolor = 220/255;

% get file list, etc
% ...

% Loop over each file in the list
%for i = 1:numel(file_list)
    % Load the image
    % ...
    
    % for sake of the demo, i'm reusing this image
    img = imread('cameraman.tif');
    
    % Get the size of the image
    [rows, cols, ~] = size(img);
    
    % Create a meshgrid of x and y coordinates
    [x, y] = meshgrid(1:cols, 1:rows);
    
    % Calculate the x and y coordinates of the center of the image
    center_x = (cols+1)/2;
    center_y = (rows+1)/2;
    
    % Calculate the radius of the oval
    radius_x = center_x - 1 - blurwidth/2;
    radius_y = center_y - 1 - blurwidth/2;
    
    % Create a mask for the oval using the x and y coordinates, center, and radii
    mask = ((x-center_x)/radius_x).^2 + ((y-center_y)/radius_y).^2 <= 1;
    
    % soften the mask
    mask = imgaussfilt(double(mask),blurwidth/4);
    
    % Create a copy of the original image with the contour region masked out
    masked_img = im2double(img).*mask + permute(bgcolor,[1 3 2]).*(1-mask);
    masked_img = im2uint8(masked_img);
    
    % show it for sake of the demo
    imshow(masked_img)

    % write the file
    % ...
%end

[Neuro]

@DGM Thanks! This worked wonderfully :)

Answer 2

[Matt J]

1.Make a logical mask of the oval

2. Erode the mask by 1 pixel (imerode)

3. Use imgaussfilt to filter the whole image

4. Restore the pixels in the interior of the oval using the mask from step 2

Answer 3

[Image Analyst]

Use imgaussfilt on the whole image. Then get an annular mask that is just the edges of the oval. Replace the pixels in that annular mask region of the original iamge with the blurred pixels. Let me know if you can't figure it out.