How to find the minimum area bounding box of a region returned from regionprops

For an image containing rectangles that are rotated with respect to the image axes, regionprops will return the bounding box coordinates of an ROI containg a rectangle, but with the ROI edges parallel with the image edges. The minimum area bounding box would be an ROI having verticies congruent with the rectangles corners; a bounding box with edges not parallel to the image axes.

Is there a function that utilizes the output of regionprops for a given region in an image and produces the minimum area bounding box? It seems this should be possible using the Feret dimensions.

Answer 1

[Image Analyst]

Yes it is. Did you actually try to use the bwferet function? If not, you should.

[Jeremy]

Yes. However, the "A suite of minimal bounding objects" at https://www.mathworks.com/matlabcentral/fileexchange/34767-a-suite-of-minimal-bounding-objects?s_tid=srchtitle was necessary to provide a "tilted rectangle" bounding box.

Even so, it seems that without the MSER "minimum diversity" parameter not being accessible in the detectMSERFeatures causes minor noise on the edges of a real rectangle (similar to a "hanging chad") in an image to be fit within the corners identified for the tilted rectangle bounding box. This could probably be mitigated for "known tilted rectangles" by performing a least squares fit on each edge of the pixel region, but the minimum diversity should allow for minor noises to be droped from the pixel list for a MSER feature.

Is the minimum diversity able to be set is some other fashion?

[Image Analyst]

I don't know. You can try editing the function - sometimes the source code is available. Then you could make a copy and code in your custom modifications.