How can I save a heterogeneous group of segmentation seeds drawn on an image via imfreehand?

I want to specify various segmentation seeds across different MRI slices and save them.

Edit: like this <http://edwardkim.net/images/3dcellular.jpg>

I'm using imfreehand to specify these seeds.

After looking at the documentation, I've thought about using roicolor, so I would get various masks based on the colors that I get on imfreehand, as seeds on different spatial locations need to be considered equal if they share a color.

I'm not sure on how to go about this cause I don't know the numerical values of the colors specified on imfreehand (which are medium blue, light blue, light red, green, yellow, magenta, cyan, light gray, black).

Anyone knows how do I find out the numerical values of these colors?

Is the idea of using roicolor sensible for my needs?

Answer 1

[Image Analyst]

roicolor() is fine for getting a binary mask of a grayscale image between a lower and an upper gray level. The color used to draw with in imfreehand() has nothing to do with anything, other than what color you see your path drawn with. Don't worry about it. Just take the values of some region you draw and from those, determine some upper and lower gray levels and use those in roicolor(). Let me know if you need a demo for how to find intensity stats for the region you drew.

[Image Analyst]

You'll probably ask for it and others might like it, so here it is:

% Demo to have the user freehand draw an irregular shape over
% a gray scale image, have it extract only that part to a new image,
% and to calculate the mean intensity value of the image within that shape.
% Also calculates the perimeter, centroid, and center of mass (weighted centroid).

% Change the current folder to the folder of this m-file.
if(~isdeployed)
  cd(fileparts(which(mfilename)));
end
clc;  % Clear command window.
clear;  % Delete all variables.
close all;  % Close all figure windows except those created by imtool.
imtool close all;  % Close all figure windows created by imtool.
workspace;  % Make sure the workspace panel is showing.
fontSize = 16;

% Read in a standard MATLAB gray scale demo image.
folder = fullfile(matlabroot, '\toolbox\images\imdemos');
baseFileName = 'cameraman.tif';
% Get the full filename, with path prepended.
fullFileName = fullfile(folder, baseFileName);
% Check if file exists.
if ~exist(fullFileName, 'file')
  % File doesn't exist -- didn't find it there.  Check the search path for it.
  fullFileName = baseFileName; % No path this time.
  if ~exist(fullFileName, 'file')
    % Still didn't find it.  Alert user.
    errorMessage = sprintf('Error: %s does not exist in the search path folders.', fullFileName);
    uiwait(warndlg(errorMessage));
    return;
  end
end
grayImage = imread(fullFileName);
imshow(grayImage, []);
axis on;
title('Original Grayscale Image', 'FontSize', fontSize);
set(gcf, 'Position', get(0,'Screensize')); % Maximize figure.
message = sprintf('Left click and hold to begin drawing.\nSimply lift the mouse button to finish');
uiwait(msgbox(message));
hFH = imfreehand();
% Create a binary image ("mask") from the ROI object.
binaryImage = hFH.createMask();
xy = hFH.getPosition;

% Now make it smaller so we can show more images.
subplot(2, 3, 1);
imshow(grayImage, []);
axis on;
drawnow;
title('Original Grayscale Image', 'FontSize', fontSize);

% Display the freehand mask.
subplot(2, 3, 2);
imshow(binaryImage);
axis on;
title('Binary mask of the region', 'FontSize', fontSize);

% Label the binary image and computer the centroid and center of mass.
labeledImage = bwlabel(binaryImage);
measurements = regionprops(binaryImage, grayImage, ...
    'area', 'Centroid', 'WeightedCentroid', 'Perimeter');
area = measurements.Area
centroid = measurements.Centroid
centerOfMass = measurements.WeightedCentroid
perimeter = measurements.Perimeter

% Calculate the area, in pixels, that they drew.
numberOfPixels1 = sum(binaryImage(:))
% Another way to calculate it that takes fractional pixels into account.
numberOfPixels2 = bwarea(binaryImage)

% Get coordinates of the boundary of the freehand drawn region.
structBoundaries = bwboundaries(binaryImage);
xy=structBoundaries{1}; % Get n by 2 array of x,y coordinates.
x = xy(:, 2); % Columns.
y = xy(:, 1); % Rows.
subplot(2, 3, 1); % Plot over original image.
hold on; % Don't blow away the image.
plot(x, y, 'LineWidth', 2);
drawnow; % Force it to draw immediately.

% Burn line into image by setting it to 255 wherever the mask is true.
burnedImage = grayImage;
burnedImage(binaryImage) = 255;
% Display the image with the mask "burned in."
subplot(2, 3, 3);
imshow(burnedImage);
axis on;
caption = sprintf('New image with\nmask burned into image');
title(caption, 'FontSize', fontSize);

% Mask the image and display it.
% Will keep only the part of the image that's inside the mask, zero outside mask.
blackMaskedImage = grayImage;
blackMaskedImage(~binaryImage) = 0;
subplot(2, 3, 4);
imshow(blackMaskedImage);
axis on;
title('Masked Outside Region', 'FontSize', fontSize);
% Calculate the mean
meanGL = mean(blackMaskedImage(binaryImage));

% Put up crosses at the centriod and center of mass
hold on;
plot(centroid(1), centroid(2), 'r+', 'MarkerSize', 30, 'LineWidth', 2);
plot(centerOfMass(1), centerOfMass(2), 'g+', 'MarkerSize', 20, 'LineWidth', 2);

% Now do the same but blacken inside the region.
insideMasked = grayImage;
insideMasked(binaryImage) = 0;
subplot(2, 3, 5);
imshow(insideMasked);
axis on;
title('Masked Inside Region', 'FontSize', fontSize);

% Now crop the image.
leftColumn = min(x);
rightColumn = max(x);
topLine = min(y);
bottomLine = max(y);
width = rightColumn - leftColumn + 1;
height = bottomLine - topLine + 1;
croppedImage = imcrop(blackMaskedImage, [leftColumn, topLine, width, height]);
% Display cropped image.
subplot(2, 3, 6);
imshow(croppedImage);
axis on;
title('Cropped Image', 'FontSize', fontSize);

% Put up crosses at the centriod and center of mass
hold on;
plot(centroid(1)-leftColumn, centroid(2)-topLine, 'r+', 'MarkerSize', 30, 'LineWidth', 2);
plot(centerOfMass(1)-leftColumn, centerOfMass(2)-topLine, 'g+', 'MarkerSize', 20, 'LineWidth', 2);

% Report results.
message = sprintf('Mean value within drawn area = %.3f\nNumber of pixels = %d\nArea in pixels = %.2f\nperimeter = %.2f\nCentroid at (x,y) = (%.1f, %.1f)\nCenter of Mass at (x,y) = (%.1f, %.1f)\nRed crosshairs at centroid.\nGreen crosshairs at center of mass.', ...
meanGL, numberOfPixels1, numberOfPixels2, perimeter, ...
centroid(1), centroid(2), centerOfMass(1), centerOfMass(2));
msgbox(message);

[Antonio]

The colors on imfreehand are important for me, cause each one specifies a different segmentation seed and different regions can be marked with same seed on an image with various kinds of seeds, (i.e. I may have 3 yellow seeds, 2 black seeds, 1 light blue seed, 4 light red seeds on the same mri slice). See?

I don't need to extract the regions, I need to save them to implement this http://www.computer.org/csdl/proceedings/cbms/2010/9167/00/06042647-abs.html across several mri slices

[Image Analyst]

There are no pictures there in your link so I don't know what you want to do. I still believe that colors are not needed for roicolor. What color you draw with has no bearing on what gray levels you specify in the gray level image. Draw with whatever color you want - it doesn't matter. Then, if you want, you can display the coordinates you drew with plot() in the overlay above your image, again in whatever color you want because it doesn't matter since the color is just in the overlay. roicolor() pays no attention to what graphics are in the overlay or their colors, it only looks at your gray level image. You can draw with 4 different colors if you want - that's fine. But when you call roicolor() you need to pass it gray levels from your original image, not colors.

[Antonio]

Look at these http://www.youtube.com/watch?v=6QhN7gP_IoM

              <http://www.youtube.com/watch?v=B997K0dQKmA>

That's what I'm implementing, look at how the user draws the seeds on the mri slices, on those demos only 2 kind of seeds (red and green) are used, but growcut can run on N number of seeds. Look at this http://en.wikipedia.org/wiki/GrowCut_algorithm and this http://graphics.cs.msu.ru/en/publications/text/gc2005vk.pdf

The point of using colors is that seeds from the same class need to be grouped together but disjointed position across several mri slices or the same mri slice.

I can ask the user to draw sequentially the seeds, and use the coordinates directly, like you say, but I feel this is cumbersome when using various slices and more than 2 segmentation classes.

[Image Analyst]

Thanks for the background. This just confirms my answer that the colors you draw with don't matter, they're just so you can tell what regions are in the "seed" region and which are in a different seed region. They could be the same color, or different colors, and whatever color they might be is totally unrelated to anything in the image. It's the color of pixels in the image under where you drew that is important, not what color you used to draw/indicate where those pixels were.

[Antonio]

Yes, I know that, the colors are "just so you can tell what regions are in the "seed" region and which are in a different seed region." I perfectly understand the segmentation algorithm, I have implemented it in C before. Now, do you understand that my problem is grouping seed regions together?

A shared color can be used to put into the same group seeds that were drawn separately across various mri slices or the same slice.

That's why I care about the colors I draw with.

[Image Analyst]

Yes, but when you draw, with whatever color, you store the drawn coordinates in the cell array for that type of region. For example you're drawing brain (in the "brain" color, whatever that may be), and your user drew three paths. So you put the first path drawn into cell array caBrain{1}, and you put the coordinates of the second path that they draw into caBrain{2}, and so on. Now the user will draw liver. So you take the first path and store that in caLiver{1}, and store the second path coordinates in caLiver{2}, and so on - and we don't care what color you use to draw liver with. You can do that because you must know what type of tissue the user is drawing over, otherwise there's no telling what they drew over. Even your references know if their user is drawing over the subject/foreground or the background, so you can also. For example, you have two buttons "Draw Brain" and "Draw Liver" and they click one, and in the callback of "Draw Brain" you assign cells of caBrain with user-drawn paths, and when they click "Draw Liver" in the callback of "Draw Liver" you assign cells of caLiver with user-drawn paths.

[Antonio]

yes, I know I can do that, as I said to you above "I can ask the user to draw sequentially the seeds, and use the coordinates directly, but I feel this is cumbersome when using various slices and more than 2 segmentation classes."

This is cumbersome and wasteful of the capacities of the algorithm, because I'll have to list a number of segmentation classes beforehand.

GrowCut has the amazing capability of performing good segmentations with N number of classes (not just foreground/background) without increasing the computation time when increasing the number of classes (unlike things like Random Walker and GraphCuts).

Having a preset number of classes is wasteful of this capability (which I want to evaluate).

Anyway, I found this and I'll try it to see if it fits what I want to do http://www.mathworks.com/matlabcentral/fileexchange/34890-fhroi-interactive-freehand-roi

If it doesn't, I'll go with some preset number of classes.