Building an outer bounding box from multiple smaller boxes

Question

I am trying to draw an outer bounding box, around all smaller bounding boxes that are overlapping. There may be many of these regions within the entire image.

e.g.

So far I have my vector of rectangles called rects.

overlaps = rectint(rects, rects); 

Where I check of overlaps with each other, and because it will compare with itself I remove the diagonal as follows:

overlaps(logical(eye(size(overlaps)))) = 0;

Then find the locations of the overlaps

[r,c] = find(overlaps > 0);

However, I am not sure how to deal with this as it is not a simple bi directional mapping in the square matrix returned, as there can be multiple overlaps in the area.

Any suggestions on how I can proceed would be greatly appreciated.

Thanks

Answer 1

here's and example with some random rectangles:

% Generate fake data, 3 rects with format [x,y,w,h]:
rects=20+randi(60,3,4);

% plot the rects :
for n=1:size(rects,1) 
    rectangle('Position',rects(n,:));
end

% get min max
xmin=min(rects(:,1));
ymin=min(rects(:,2));
xmax=max(rects(:,1)+rects(:,3));
ymax=max(rects(:,2)+rects(:,4));

% define outer rect:
outer_rect=[xmin ymin xmax-xmin ymax-ymin];

hold on
rectangle('Position',outer_rect,'EdgeColor','r','LineStyle',':');

Answer 2

function localBoxes = maxBoxVals(finalBoxes)

rects = [];
for es = 1 : length(finalBoxes)
y = finalBoxes(es, 1); 
x = finalBoxes(es, 2); 
y2 = finalBoxes(es, 3);
x2 = finalBoxes(es, 4); 
rects = [rects ; [x, y, y2-y, x2 - x]];
end

overlaps = rectint(rects, rects);

%overlaps(logical(eye(size(overlaps)))) = 0;

 [r,c] = find(overlaps > 0);
 localBoxes = [];


for i = 1 : length(overlaps)
col = overlaps(:,i);
col2 = find(col > 0);

if(~length(col2) > 0)
    continue;
end
localRects = [rects(i,:)];
for j = 1 : length(col2)
    localRects = [localRects ; rects(col2(j),:)];
end

if(max(localRects) > 0)
    minX = min(localRects(:,1));
    minY = min(localRects(:,2));
    maxX = max(localRects(:,1))+max(localRects(:,3));
    maxY = max(localRects(:,2))+max(localRects(:,4));
    localBoxes = [localBoxes ; [minX,minY,abs(maxX - minX), abs(maxY - minY)]];
end

end

localBoxes = unique(localBoxes,'rows')



 end

This is what I have come up with to solve. Not vectorised and definitely not optimal, and may be some bugs, but seems to get the job done for what I need, and applies to the whole image.