How to automatically delete part of the mask like this

Hi dear all,

I am working on a segmentation methods using region growing. The segmented results always leak out a little bit (marked as red). I do know the starting seeds location and looking for a way that can help me get rid of the rest of the mask. They do have some features but I cannot come up with a good way to do the task.

Answer 1

[Image Analyst]

It would help to see the original gray scale images to know why your segmentation was not successful. It might be easier to prevent the "overflow" than to remove it once it's there.

Anyway I'd experiment around with imopen or watershed to try to separate the blobs. You might also need bwdist and/or imreconstruct

[Haowei]

Hi thanks for getting back to me. I think the issue from the image was that the components are connected like tubes, however, I just need one part of it (e.g., the black region on the middle left is what I need and also the bottom black also needs to be excluded. Basically ends where the two bright stripe faints). I might use machine learning to deal with these cases.

[Image Analyst]

For deep learning, like SegNet or Unet you'd need to provide hundreds of images where you've manually outlined ("labeled") the regions you want it to find. If you're going to do that and spend the time to write a training program and program to apply the network, you might as well just hand draw the regions you want. Either way you're hand-drawing regions.

I'm attaching some hand drawing masking demos.

[Haowei]

Thank you so much. I agree that making labels is a pain but region growing is hard to disconnect those parts (that is also a pain). Sincerely appreciate your help, thanks again for your support.