Image classification - coloring features with a colormap? - matlab

Question

I have an image with some features/regions in it (balls in the above example). I want to color each ball with a different color based on its properties. For example, that might be its diameter in pixels.
While I'm done on the feature recognition side, I'm stuck when it comes to showing results. Right now I'm doing:
my_image = imread(...);
//ball recognition and other stuff
for i = 1:number_of_balls
ball_diameter(i) = ... //I calculate the diameter of the i-th ball
ball_indices = ... //I get the linear indices of the i-th ball
//ball coloring
my_image(ball_indices) = 255; //color the red channel
my_image(ball_indices + R*C) = 0; //color the blue channel
my_image(ball_indices + 2*R*C) = 0; //color the green channel
end
figure
imshow(my_image)
colormap jet(100) //I want 100 different classes
colorbar
caxis([0, 50]) //I assume all balls have a diameter < 50
In the above code I'm tinting all balls red, which is definitely not what I'm looking for. The issue is that, even if I know ball_diameter(i), I do not know which colormap class that ball will get in. In other words, I would need something like:
for i = 1:number_of_balls
// ...
if ball_diameter *belongs to class k*
my_image(ball_indices) = jet(k, 1);
my_image(ball_indices + R*C) = jet(k,2);
my_image(ball_indices + 2*R*C) = jet(k,3);
end
end
How to, and mostly, is there any other more logical way?

Answer 1

You can separate the assignment of pixels to classes from their coloring for display: you can use my_image as a 2D R-by-C labeling matrix: that is each object (ball) is assigned a different index from 1 to 100 (in case you have 100 objects in the image). Now when you want to display the result you can either ask the figure to map indexes to colors for you, using colormap or explicitly create a colored image using ind2rgb.
For example
%// create the index/labeling matrix
my_image = zeros( R, C );
for ii = 1:number_of_balls
my_image( ball_indices ) = ii; %// assign index to pixels and not atual colors
end
%// color using figure
figure;
imagesc( my_image );axis image;
colormap rand(100,3); %// map indexes to colors using random mapping
%//explicitly create a color image using ind2rgb
my_color_image = ind2rgb( my_image, rand(100,3) );
figure;
imshow( my_color_image ); % display the color image
Notes:
1. IMHO it is preferable to use random color map to display categorizations of pixels, as opposed to jet with which you usually end up with very similar colors to adjacent objects, making it very difficult to visually appreciate the result.
2. IMHO it is more convenient to use label matrix, you can also save it to file as an indexed image (png format) thus visualizing, saving and loading your results simply and efficiently.

Answer 2

A simple way is as follows:
Make a 2D image of the same size as your original.
Set the indices of each "ball" to the diameter or other relevant value
Display with imagesc
Use caxis, colormap, colorbar etc. to adjust the categories dynamically.
For example,
a = randi(200,200); % 200 x 200 image containing values 1 to 200 at random
imagesc(a)
colorbar
The above should show a random color field with the default colormap. The colorbar goes from 1 to 200.
colormap(jet(5))
The colorbar still goes from 1 to 200, but with only 5 colors.
caxis([1 100])
The colorbar now shows the five colors scaled from 1 to 100 (with everything above 100 in the top pot).
If you want to convert your 2D image full of different diameters to a set of discrete labels indicating diameter ranges, an easy way is to use histc with the second output bin being the same size as your input image, set to which bin the diameter value fell into. The second input in this case is the edges of the bins, not the centres.
[n bin] = histc(a,0:20:201);