Why VeriLook?
Click to show
multiple faces
detection |
Neurotechnologija has developed a PC-based face recognition algorithm VeriLook 3.2 designed for biometrical system integrators. VeriLook 3.2 offers capabilities of the most advanced and convenient facial identification systems at a reasonable cost:
- Reliability. VeriLook 3.2 algorithm has been tested with standard face databases (FERET, XM2VTSDB and others). The results are one of the best among existing face identification systems on the market.
- Speed. VeriLook 3.2 face enrollment time is less than 0.3 sec and matching speed is 100,000 faces per second in 1:N identification mode.
- Live face detection. A conventional face
identification system can be easily cheated by placing a photo of
another person in front of a camera. VeriLook is able to prevent this
kind of security breach by determining whether a face in a video stream
belongs to a real human or is a photo.
- Multiple face processing. VeriLook 3.2 detects all faces in the current frame and allows to process all of them simultaneously.
- VeriLook doesn't require any specific hardware. Face image can be obtained from a webcam or other low cost camera. Image processing and recognition are performed on ordinary PC.
-
Biometrical template record can contain multiple face samples belonging to the same person.
These samples can be enrolled with different face postures and
expressions, from different sources and in different time thus allowing
to improve matching quality. For example a person could be enrolled
with and without eyeglasses or with different eyeglasses, with and
without beard or moustache, smiling and non-smiling etc.
- Both face and fingerprint recognition technologies from the same vendor. Compatible product interfaces and customer support from the same source allow simple multi-biometric system integration and helps to achieve high system recognition quality. VeriLook algorithm can be used alone or together with other Neurotechnologija biometrical algorithms.
- VeriLook is designed not only for verification, but also for identification (1:N matching).
|
|
Algorithm
VeriLook 3.2 face recognition algorithm implements advanced face localization, enrollment and matching using robust digital image processing algorithms:
-
Fast and accurate face localization for reliable detection of multiple faces in live video streams and still images.
-
Simultaneous multiple face processing and identification in single frame.
All faces on the current frame are detected in 0.07 sec.* and then each face is processed in 0.13 sec.*.
-
Face quality threshold can be used during face enrollment to ensure that only the best quality face template will be stored into database.
-
The VeriLook 3.2 face template matching algorithm compares 100,000 faces per second*.
-
Applications implemented using VeriLook SDK can handle large face databases, as one face features template is only 2.3 Kbytes.
-
Features generalization mode
generates the collection of the generalized face features from several
images of the same subject. Then, each face image is processed,
features are extracted, and the collections of features are analyzed
and combined into a single generalized features collection, which is
written to the database. This way, the enrolled feature template is
more reliable and the face recognition quality increases considerably.
-
VeriLook has certain tolerance to face posture
that assures face enrollment convenience: rotation of a head can be up
to 10 degrees from frontal in each direction (nodded up/down, rotated
left/right, tilted left/right).
-
Biometrical template record can contain several face samples belonging to the same person.
These samples can be enrolled from different sources and in different
time thus allowing to improve matching quality. For example a person
could be enrolled with and without eyeglasses or with different
eyeglasses, with and without beard or moustache, etc.
Note: All performance evaluations were performed using PC with 3 Ghz Intel Pentium4 CPU.
Reliability Tests
Click to enlarge |
|
VeriLook 3.2 was tested with face images from many cameras. The most interesting testing results are those obtained with standard databases, because in this case they can be compared with testing results of other algorithms. Usually the algorithm recognition quality is expressed by receiver operating curves (ROC), which show the dependence of false rejection rate on the false acceptance rate. The presented ROC shows the results of testing VeriLook 3.2 with face images from XM2VTSDB database. The red curve shows the performance results of VeriLook 3.2 without generalization, and the green one shows the results of VeriLook 3.2 with generalization.
|
| VeriLook 3.2 algorithm technical specifications |
| Recommended minimal image size |
640 x 480 pixels |
Multiple faces detection time
(using 640 x 480 image) |
0.07 sec. |
Single face processing time
(after detecting all faces) |
0.13 sec. |
| Matching speed |
100,000 faces/sec. |
| Size of one record in the database |
2.3 Kbytes |
| Maximum database size |
unlimited |
All performance evaluations were performed using a PC with 2.8 GHz Intel Pentium4 CPU
|
System Requirements
PC with Pentium 1GHz processor;
128 Mb of RAM;
Related products:
These products are based on VeriLook algorithm:
|
