How to threshold an image when background and foreground color are similar to each other?

Hi i would like to mask the background in this image so i have only the face of this baby as the output. But the background colors are very similar to the face skin color so when I'm applying thresholding im also losing the portion of the face or including the region that are unnecesary. How can I tackle this issue? What pre processing techniques should use? Or should take a different approach to the problem? What would you suggest?

Thanks in advance.

Answer 1

[yanqi liu]

clc; clear all; close all;
img = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/819834/image.png');
% method 1
jmg = rgb2ycbcr(img);
s = jmg(:,:,2);
bw = ~imbinarize(s);
bw = bwareafilt(imopen(bw, strel('disk', round(size(bw,2)/5))), 1);
stats = regionprops(bw);
figure;
subplot(2, 2, 1); imshow(img);
subplot(2, 2, 2); imshow(s, []);
subplot(2, 2, 3); imshow(bw, []);
subplot(2, 2, 4); imshow(img, []);
hold on; rectangle('position', stats(1).BoundingBox, 'EdgeColor', 'g', 'LineWidth', 2)

% method 2
faceDetector = vision.CascadeObjectDetector();
faceDetector.ScaleFactor = 1.01;
bbox            = step(faceDetector, img);
figure;
imshow(img);
hold on; rectangle('position', bbox(1,:), 'EdgeColor', 'g', 'LineWidth', 2)

[Oguzhan Guvercin]

Thank you for your answer. Your first method was the answer that I was looking for. Cause this is for a Digital Signal Processing course project and I am not allowed to use any classifier algorithm whatsoever.

By the way if you have the time can you explain your first method for me cause i would like to learn your thinking process when you encounter a problem like this. If you could do that that would mean a lot for me.

But either way again thank you so much.

[yanqi liu]

yes，sir

The main idea is to transform the color space and extract the skin color region. The face has obvious continuous skin color characteristics, and there are face edges around. Therefore, although the background is partially repeated, the face is still a relatively large region. Through morphological filtering, the interference of the background can be eliminated and the face candidate region can be obtained. Of course, other color space analysis can also be considered.

[Oguzhan Guvercin]

Thank you sincerely sir. I appreciate for your reply and time.