Tips for 3D Face Scanning with Hair
Posted by Thomas Tong on Wed, Feb 13, 2013 @ 11:00 AM
In the past, capturing hair was an exercise in futility. For a head scan, the facial features would look great, but the person's hair would barely be visible or it would not show up at all. Scanning hair can be a challenge for many types of 3D scanners including structured-light systems. Hair consists of very fine and partially translucent filaments which scatter the light and are often interpreted as noise during 3D reconstruction. As a result, these areas are usually removed from typical scan data. With the release of the latest version of FlexScan3D software (version 3.1.7) that powers the HDI 3D Scanners, we introduced a new High Sensitivity mode to address this shortcoming.
This article provides an overview on acquiring usable hair scans specifically for FlexScan3D. While it is mainly targeted at scanning human hair, it also applies to scanning other objects, such as fuzzy clothing, stuffed animals, or dense vegetation.
3D Scanning Process
Since the hair and face (skin) properties differ drastically, our team found that processing them separately using different settings and then combining the results later generated the best results.
Here are the general steps:
- Capture 3D scans of the subject to create a 3D mesh
- Process the mesh to optimize for the hair
- Duplicate the mesh and process it to optimize for the face
- Combine the hair and face mesh together to create the final 3D model
Here is an illustrated version of these steps:
1) Several raw 3D scans are aligned together to create a 3D mesh
2) The left mesh is processed and optimized for the face/skin while the right mesh is processed and optimized for the hair
3) The hair mesh is edited by taking out the face
3) Combine the hair and face mesh together
Final 3D Model
Tip for Aligning 3D Scans: Since hair scans are usually quite rough, geometry-based alignment may be unreliable for 3D scans mainly consisting of hair. Geometry-based alignment uses a specific feature in the 3D scans as a point of reference to align them together. Instead of relying purely on the rough hair scan data, first ensure that each scan contains other surface information such as a facial or body feature. For body scans, this can simply be the person's torso.
For scanning an object which has no smooth surfaces, such as a teddy bear or a fuzzy sweater, easily-scannable items (ones with smooth and opaque surfaces) should be strategically placed around the object. Geometry alignment can then be done by selecting the geometry of these external items.
Here is a setup of a teddy bear (scan object) with items surrounding it to make aligning the 3D scans much easier. Reference items can be deleted after alignment and merging is complete. Using a rotary table is helpful for 360 degree scanning to scan all sides while minimizing the movement of objects.
High Contrast Scans: For high-contrast scans (light-skinned with dark hair or dark-skinned with light hair), the camera exposure used for capturing the skin will probably not work for capturing the hair. In these cases, either use FlexScan3D's HDR-mode scanning, or capture the skin and the hair separately, then merge the results afterwards.
Accuracy: While High Sensitivity mode is very useful for scanning hair, local accuracy is not guaranteed when it is enabled. The 3D reconstruction algorithm smooths over holes and cracks in the source data, and may fill in certain areas erroneously. As a result, it is not recommended for scanning typical surfaces.
For more detailed instructions on this process, please refer to the user manual.
Have questions about face or hair scanning? Please comment and share your thoughts.