How do I automatically detect an object and create its BoundingBox?

Hello, I want to detect the pill package on the table automatically in my script and then use the BoudingBox to crop the image. So far, I am working around this problem by automatically cropping the image (displayed as green box in image). The detection doesn't need to be 100% exact, it can be more or less like the green box in the image.

I already tried adaptive thresholding and multilevel thresholding, but it does seem to work and I get really bad results which are unusable. Can anybody help me?

The multilevel thresholding of the Saturation channel of this image looks like this:

During the script I also create an image based on the local standard deviation, which looks like the image below. It seems that it could somehow be possible to detect the borders of the pill package but I couldn't figure out how...

[Walter Roberson]

What happens if you first threshold for brighter white, and then bwareafilt() to remove what speckle noise remains? I suspect that the resulting image would have gaps but you would still be able to do boundary detection on it.

[Jannis Holtkoetter]

Hey, thank you, it really does a good job so far. I used some more morphological operations and the results are quite okay.

However, on some images there can be some brighter pixels in the background...then the binary image looks like this:

Does anybody know, how to remove the large white blob, which is attached to the blister?

[Shantanu Dwivedi]

Have you tried the following edge detection techniques ?

https://www.mathworks.com/discovery/edge-detection.html