Fit rectangle to binary mask

I have a binary mask prediction imported into matlab from neural network training. It's just a large array filled with zeros, except where the predictied structures lies, it is filled with ones. The predicted structure looks more ore less like a rectangle and i need to get these 4 corner coordinates from the predicted mask. So my idea would be to fit a rectangle to the predicted mask and read out the corner coordinates of the rectangle. Is there an existing function for this? Or does anyone have another approach?

Answer 1

[Matt J]

See the FEX submission pgonCorners (which you must Download).

[Stefan Lang]

Is there any way to fit a rectangle to these corner points? Sometimes 2 of the proposed corner points from pgonCorners lie in the same corner region. If i ask to propose e.g. 8 points, each corner region has some corner points in it. How can i fit a rectangle to these 8 points and get the corner coordinates of the rectangle?

[Matt J]

Sometimes 2 of the proposed corner points from pgonCorners lie in the same corner region.

It would be good to see a sample image where that occurs. That shouldn't happen if pgonCorners is run with the right settings.

[Stefan Lang]

I tried it with the attached file and the following code:

vertebra_corners_frontal = corner_prediction(vertebras_frontal,4,100)

pgonCorners is inside corner_prediction. Some vertebras get 2 corner points in one corner. If i adjust angular sample, i even sometimes get only 3 points for each region.

[Matt J]

Again, it would be good to see a sample image where that occurs.