Face anti-spoofing countermeasure: Efficient 2D materials classification using polarization imaging
Spoofing is an act to impersonate a valid user of any biometric systems in order to gain access. In a face biometric system, an imposter might use some fake masks that mimic the real user face. Existing countermeasures against spoofing adopt face texture analysis, motion detection and surface re...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | English |
| Published: |
2017
|
| Subjects: | |
| Online Access: | http://eprints.unisza.edu.my/787/1/FH03-FIK-18-15467.pdf http://eprints.unisza.edu.my/787/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Spoofing is an act to impersonate a valid user of
any biometric systems in order to gain access. In a face biometric
system, an imposter might use some fake masks that mimic the
real user face. Existing countermeasures against spoofing adopt
face texture analysis, motion detection and surface reflection
analysis. For the purpose of face anti-spoofing analysis, skin
structure is a key factor in achieving the target of our study. Skin
consists of multiple layers structure which produces multiple
reflections: surface and subsurface reflections. In this paper, we
proposed a measure to discriminate between a genuine face and a
printed paper photo based on physical properties of the materials
which contribute to its distinctive reflection values. In order to
differentiate the reflections, polarized light (light that vibrates in
a single direction) can be used. The Stokes parameters are
applied to generate the Stokes images which are then used to
produce the final image known as Stokes degree of linear
polarization (SDOLP) image. The intensity of the SDOLP image
is investigated statistically which has shown promising results in
the materials classification, between the skin and the paper mask.
Furthermore, comparison between the experimental results from
two skin color groups, black and others show that the SDOLP
data distribution of black skin is similar to the printed paper
photo of the same skin group. |
|---|