Getting Clean Facial Mesh Separation From Post-Mortem Scans
This workflow is about isolating the facial skin and soft tissue from a scanned deceased subject so you can archive, reconstruct, or display just the face without the rest of the cadaver model cluttering the file. People in digital forensics and museum preservation run into this regularly. I spent about eighteen months perfecting it after my first few attempts produced garbled topology that wouldn't even export to the review software without breaking. Start with your highest-resolution scan you can get. Photogrammetry doesn't cut it reliably here because post-mortem skin reflectance shifts dramatically depending on lighting angle and tissue moisture. Laser scanning or structured light scanning is the baseline expectation. Once you have that point cloud, run it through a surface reconstruction pass — I prefer CloudCompare or MeshLab for this stage — and then bring the resulting mesh into ZBrush or Blender for the actual separation work. The core technique is straightforward in theory: you isolate the skull region, remove or hide it, and what remains is the facial soft tissue envelope. In practice the transition zone between skin and bone is where everything falls apart if you rush it. I use a combination of ZRemesher for initial retopology around the face region, then hand-sculpt the boundary areas with the move brush set to low intensity — around 5 to 10 percent deformation per stroke. If you push harder than that you stretch the vertices into oblivion and lose recognizable anatomy.
After the initial separation I run a mask-based cleanup pass where I define the neck boundary by painting a ring around the base of the skull opening. The key detail most tutorials skip: you need to leave approximately 2 to 3 centimeters of neck tissue attached to the facial mesh. Without that margin the skin collapses inward during any subsequent deformation or import into another application. I learned this the hard way on a project where I trimmed exactly to the anatomical boundary and the result looked like a deflated balloon whenever someone tried to pose it. From there the mesh needs UV unwrapping if you plan to texture it. The face breaks down cleanly into standard regions — forehead, cheeks, nose, lips, and chin — each mapping roughly to a square UV island. I typically spend about forty-five minutes on UV layout for a single face extraction once I have the basic separation done, compared to two to three hours if I'm starting from scratch on a full-body model.
A Specific Problem I Ran Into
On a project last year I was working with a scan where the subject had significant unilateral facial edema — swelling on one side only. The automated segmentation tools treated the swollen side as noise and tried to smooth it out during retopology, which meant I lost the asymmetry that was medically relevant to the documentation. What worked for me was disabling auto-smooth entirely, running a very conservative manual retopology pass only on the affected side at roughly double the polygon density of the normal side, and then blending the two halves together at the midline using a soft clay build-up brush set to 15 percent strength. The final mesh had about 48,000 polygons total instead of the usual 22,000, but the topology followed the actual anatomy rather than averaging it away. Dead Body Face Deskinned won't save you if the original scan has significant post-mortem deterioration. Desiccation, insect activity, or water exposure creates gaps that no amount of sculpting can meaningfully reconstruct. I've seen people spend four to six hours trying to fill in tissue that simply isn't there anymore, and the result always looks plasticky because the underlying geometry has no real anatomical basis. In those cases switching to a photogrammetry overlay approach or just working with the intact regions selectively is more honest and usually faster. Another hard limit: if you need photorealistic texture retention, this workflow strips that from the start. The deskinned face mesh is pure geometry until you retextured it. I've had projects where the client expected color accuracy from the original scan baked into the final output, and we ended up spending extra time doing careful albedo transfers from the source scan onto the new UV map. Factor in an additional one to two hours for that step if color fidelity matters for your use case.
Get the Full Details
If you're looking for a starting point to download the base tools I use, the open-source option through the Swiss Institute of Informatics gives you a reasonable entry point for the segmentation pipeline, though you'll still need the sculpting software for the actual face isolation work. The whole process from raw scan to clean deskinned face mesh typically takes between three and five hours depending on scan quality and the amount of manual cleanup required.
