{"id":5965,"date":"2015-05-08T09:00:09","date_gmt":"2015-05-08T13:00:09","guid":{"rendered":"https:\/\/blogs.mathworks.com\/pick\/?p=5965"},"modified":"2017-08-29T14:15:43","modified_gmt":"2017-08-29T18:15:43","slug":"how-do-you-create-a-mask-from-a-variable-thickness-open-freehand-curve","status":"publish","type":"post","link":"https:\/\/blogs.mathworks.com\/pick\/2015\/05\/08\/how-do-you-create-a-mask-from-a-variable-thickness-open-freehand-curve\/","title":{"rendered":"How do you create a mask from a variable-thickness open, freehand curve?"},"content":{"rendered":"

Brett<\/a>'s Pick this week is 2D Line Curvature and Normals<\/a>, by Dirk-Jan Kroon<\/a>.<\/p>

Recently, I began working on an app to facilitate interactive image segmentations. Ultimately, I want to create a suite of tools that allow me to use some combination of automated and manual manipulations to create flexibly a complicated mask of my images. As part of that project, I want the ability to create closed or open (!)<\/i> imfreehand<\/tt><\/a> or impoly<\/tt><\/a> regions, and to use those regions to modify my mask.<\/p>

Creating masks from closed regions is easy with the imroi<\/tt><\/a> tools of the Image Processing Toolbox<\/a>; the \"createMask\" method trivializes the process. But creating masks from open regions can produce some unexpected, and unwanted, results.<\/p>

Suppose, for example, that I wanted to create a mask from a manual tracing of the lower left petal of the \"yellow lily\" image that ships with the Image Processing Toolbox:<\/p>

I read in and display the image, and trace the boundary of interest:<\/p>

lily = imread('C:\\Program Files\\MATLAB\\R2015a\\toolbox\\images\\imdata\\yellowlily.jpg'<\/span>);\r\nax(1) = subplot(1,2,1);\r\nimshow(lily)\r\ntitle('Lovely Lily'<\/span>)\r\n% I did this manually, saved the positions to a Mat-file:<\/span>\r\n% h = impoly(imgca,'closed',false);<\/span>\r\nload outlinedLily<\/span>\r\nh = impoly(imgca,xsys,'closed'<\/span>,false);\r\nxsys = getPosition(h);\r\nxs = xsys(:,1);\r\nys = xsys(:,2);\r\n% (Note: you can do this with |imfreehand|, too, but I find that I can be<\/span>\r\n%  much more accurate with |impoly|.)<\/span>\r\ndefaultMask = createMask(h);\r\nhold on<\/span>\r\nplot(xs,ys,'r.'<\/span>,'markersize'<\/span>,16);\r\n<\/pre>
If I simply use the createMask method, I get a mask of the filled region within the closed ROI:<\/p>
ax(2) = subplot(1,2,2);\r\nimshow(defaultMask);\r\ntitle('Default createMask'<\/span>)\r\n<\/pre>
\"\" <\/p>
But that's not what I wanted! I could use bwperim<\/a> (for example) on that mask, but then I would have to remove the effects of the unwanted region closing. What I really<\/i> wanted was a thin \"shell\" around my drawn region, from which I could create my mask.<\/p>
Enter Dirk's LineNormals2D<\/tt>.<\/p>
N = LineNormals2D(xsys); %Dirk-Jan!!!! This rocks!<\/span>\r\nthicknessMultiplier = 2;\r\nposn = [xs-thicknessMultiplier*N(:,1) ys-thicknessMultiplier*N(:,2);\r\n\tflipud(xs+thicknessMultiplier*N(:,1)) flipud(ys+thicknessMultiplier*N(:,2))];\r\nh = impoly(ax(2),posn);\r\nxsys = getPosition(h);\r\nxs = xsys(:,1);\r\nys = xsys(:,2);\r\nplot(ax(1),xs,ys,'g.'<\/span>,'markersize'<\/span>,16);\r\n<\/pre>
\"\" <\/p>
Now...<\/p>
imshow(createMask(h),'parent'<\/span>,ax(2))\r\ntitle('That''s more like it!'<\/span>)\r\n<\/pre>
\"\" <\/p>
The line above, in which I used the output of LineNormals2D<\/tt> to create my \"posn\" vector, might seem a bit puzzling. In a nutshell, Dirk-Jan's function calculated the position of the normal<\/i> at each point on my imroi<\/tt>. I subtract those values from, and then add them to, the points on my curve to calculate desired positions. (thicknessMultiplier just scales the amount of offset.) I use flipud<\/tt><\/a> to have the \"inward points\" flip around and start where the \"outward points\" left off:<\/p>
t = 0:pi\/64:3*pi;\r\nxy = [t',sin(t)'];\r\nN = LineNormals2D(xy);\r\nplot(xy(:,1),xy(:,2),'b.'<\/span>)\r\nplot(xy(:,1)-N(:,1),xy(:,2)-N(:,2),'r.'<\/span>)\r\nplot(xy(:,1)+N(:,1),xy(:,2)+N(:,2),'g.'<\/span>)\r\nlegend({'Original points'<\/span>,'XY-N'<\/span>,'XY+N'<\/span>})\r\nxlabel('t'<\/span>)\r\nylabel('sin(t)'<\/span>)\r\n<\/pre>
\"\" <\/p>
Zooming in, we can see what LineNormals2D did more explicitly:<\/p>
h = impoly(ax(2),posn);\r\nsetColor(h,[1 0 1])\r\nax.xlim\r\nset(ax,'xlim'<\/span>,[165 230],'ylim'<\/span>,[1120 1200])\r\n<\/pre>
\"\" <\/p>
There are undoubtedly other ways of creating this mask, but this is what initially occurred to me. Swag, of course, goes to Dirk-Jan for his fine code, but also to anyone who shows me another clever way to implement this functionality!<\/p>
As always, I welcome your thoughts and comments<\/a>.<\/p>