Another technique we have already covered, that is changed in the task of color image processing, is image thresholding. Color images must be thresholded in each channel separately using a proper threshold and then the results must be combined together. Let's try to do this using the automated Otsu thresholding technique we presented in Chapter 2, Working with Pixels in Grayscale Images.

First, the color channels should be thresholded one by one. Let's see how, writing a script under the name ColorOtsuThresholding.m:

img_color = imread('my_image_color.bmp'), % Load image
red = im2bw(img_color(:,:,1)); % Threshold red channel
green = im2bw(img_color(:,:,2)); % Threshold green channel
blue = im2bw(img_color(:,:,3)); % Threshold blue channel
bin_image_or = red | green | blue; % Find union using OR
bin_image_and = red & green & blue; % Find intersection using AND
subplot(1,3,1),imshow(img_color),title('Original Image')
subplot(1,3,2),imshow(bin_image_or),title('Binary Union Image')
subplot(1,3,3),imshow(bin_image_and),title('Binary Intersection Image')

Running this script gives the following result:

Note that the union of two or more binary images can be acquired using the OR operator, which in MATLAB, is denoted by symbol "|". Applying this operator to two binary images, results in an image that contains ones in those pixels that are equal to one in at least one of the two images. The AND operator denoted by symbol "&" leads to a resulting image that contains ones in those pixels that are equal to one in both images. Depending on the task, one method could be preferable to the other. Let's try to illustrate the difference with an example.