ROI with hole, convert back to logical

Hi,

I have a task that has probably been asked before, but I do not find something about this issue.

I have a binary mask that is converted to a ROI, then edited, then converted back to a binary mask. The binary mask contains objects with holes in it. The ROI contains objects for the holes, but when converting back to binary, these holes are lost. I am looking for a simple soution for this (code is mainly from Matlab help):

A=zeros(1000,1000,'logical');
A(200:800,200:800)=1;
A(300:400,300:400)=0;
blocations = bwboundaries(A,'holes');
figure
imshow(A, []);
for ind = 1:numel(blocations)
    % Convert to x,y order.
    pos = blocations{ind};
    pos = fliplr(pos);
    % Create a freehand ROI.
    drawfreehand('Position', pos);
end


% Convert edited ROI back to masks.
hfhs = findobj(gca, 'Type', 'images.roi.Freehand');
editedMask = false(size(A));

for ind = 1:numel(hfhs)
    % Accumulate the mask from each ROI
    editedMask = editedMask | hfhs(ind).createMask();

    % Include the boundary of the ROI in the final mask.
    % Ref: https://blogs.mathworks.com/steve/2014/03/27/comparing-the-geometries-of-bwboundaries-and-poly2mask/
    % Here, we have a dense boundary, so we can take the slightly more
    % performant approach of just including the boundary pixels directly in
    % the mask.
    boundaryLocation = hfhs(ind).Position;
    bInds = sub2ind(size(A), boundaryLocation(:,2), boundaryLocation(:,1));
    editedMask(bInds) = true;
end

The aim is that "editedMask" looks like "A" after these two conversions. Thanks for your help!!

[DGM]

Can't you just do a logical combination of the derived masks?

theblob = createMask(hfhs(2));
thehole = createMask(hfhs(1));
blobwithhole = theblob & ~thehole;

imshow(~blobwithhole)

Or are you asking if there's a way to automatically determine whether an ROI represents a hole or not?

[William Thielicke]

Hi, thanks for your input and sorry for the delayed feedback. What I didnt yet mention is that I dont know much about the amount of ROIs. It might also bet that inside the hole there is another object (what is the word for the opposite of "hole"...?).

Therefore, I think the following approach might work for most cases:

editedMask = zeros(size(A),'uint16');
for ind = 1:numel(hfhs)
    editedMask = editedMask + uint16(hfhs(ind).createMask());
end
editedMask = bitget(editedMask,1);

So each region that has an uneven amount of ROI overlays becomes zero, and regions with an even amount of ROI overlays becomes one.

[Image Analyst]

What happened to your first (badly named) "A"? Why can't you just keep that in memory? How did it get cleared? Can you just not clear it? Or else pass it into whatever function needs the original "A"?

[DGM]

Will that be okay if you have two "holes" overlap or something? Should that create one large hole with an unattached island, or should it just be one large hole? I mean, these are design decisions for you, but I think your method is one possible solution.

[William Thielicke]

@Image Analyst: These "masks" are meant to be used in a GUI to exclude certain regions of images from analysis. These masks can be generated by the user in different ways: Drawing by hand, or importing a user-generated binary image. In both cases, I want to keep the masks editable until they are used in the final image analysis. I also want to offer some basic image processing on the imported binary mask: imdilate, imerode, imclose etc. Therefore, I think the best way is to work with Matlabs ROI objects all the time, and convert them back to binary at the very end (because a binary mask is what the final analysis expects).

I am trying to find a good solution that really helps people to define masks, but as @DGM noticed, I have to take a lot of design decisions which feels pretty difficult at the moment.

Currently, the even/uneven method seems to be compatible with imported binary masks and with masks that are drawn with polygons. I personally think the behaviour makes sense, but it is difficult to assess how others feel about it.

Polygon mask behaviour:

User input

Conversion to pixels result:

Pixel mask behaviour:

User input (binary pixel image):

Conversion from pixel mask to polygon mask (editable):

Conversion back to pixels result: