search

SCIInstitute / Seg3D

Seg3D is a free volume segmentation and processing tool developed by the NIH Center for Integrative Biomedical Computing at the University of Utah Scientific Computing and Imaging (SCI) Institute.
110 stars 47 forks source link

Seed point disconnect region filter #57

Open a-y-khan opened 9 years ago

a-y-khan commented 9 years ago

Request from Paul Holcomb through mailing list:

EM images, especially of neural tissue, are dense, with many objects packed closely together. Usually, with thresholding and a connected components filter I can tease out large objects. However, when it comes to following small processes, this approach breaks down quickly. Additionally, the size of the processes and the inherent noise in the images makes eroding until the desired part of a mask is essentially disconnected impossible. Usually, what I end up with is a good approximation of the process but still connected to adjacent processes. It would be useful to have a tool where I could place two seed points, one in the part of the mask I want to keep and the other that should be disconnected. The tool would then find the pixels closest to the two seed points that are also adjacent to one another, and then delete the pixel closest to the seed point of the part of the mask to be thrown away. This process would iterate until the portion of the mask containing the unwanted seed point is disconnected, and then remove that portion of the mask. I don't know if that makes any sense, or if it would be difficult to implement, but it would speed segmentation of these small processes by at least an order of magnitude.

a-y-khan commented 9 years ago

Maybe adapting the connected component filter would work? Will have to dig into the algorithm.

jessdtate commented 5 years ago

this is similar to segmentation level set filter, which already exist. It is more similar to the grow-cut in core view.