Face Recognition Systems

TEAM-1JACKIE ABBAZIO

SASHA PEREZDENISE SILVA

ROBERT TESORIERO

Face Recognition Systems

Overview: Face Biometrics

Facial recognition through the use of computer analysis of facial structure.

Software measures a number of points of facial characteristics such as eyes, nose, mouth, angles of key features, and lengths of various portions.

Collected data is used to create a template through a mathematical algorithm (Neural-Networks, Eigenface) and the file is stored within the database.

File is compared to other files within the database in search of an identity match.

Face Biometrics Continued

Face biometric systems employee the capturing of facial pattern characteristics through the use of still photography or video clips.

Pattern recognition software relies on: Data collection – raw data Feature extraction – eyes, nose, mouth, etc Classification – class the object is placed into (male or

female, skin tone, etc)

FAR & FRR

FAR ( False Acceptance Rating) – the false acceptance rating is the probability that the software will incorrectly declare a successful match between the input data against the database.

FRR (False Rejection Rating) – the false rejection rating is the probability that the software will declare a failure to match the input data against the database.

Face Biometric Systems Project

The face biometric systems project involved the research and testing of various facial biometric software’s based on criteria set by the clients.

Two face biometric software’s were chosen and tested (Luxand FaceSDK & VeriLook).

Software Comparison Table

Luxand’s FaceSDK 1.7

After extensive testing and researching the Face Biometric Systems, we recommend purchasing

Luxand’s FaceSDK 1.7 software. It has the following strengths: Easy to use Ability to enroll all images Matches work best at FAR of 50%, but produces

matches at FAR of 10%. Works best for aging

*Free demo version used

Enrolled into class databaseEnrolled into class database

This 1979 image matched with 2007 image2008 image with 51% similarity

This 1979 image matched with 2007 image2008 image with 51% similarity

Luxand FaceSDK Example

Luxand FaceSDK Similarity Matrix

VeriLook 3.2

Designed for biometric system developers and integrators.

Allows for easy integration and rapid development of biometric applications using functionality.

Can perform simultaneous multiple face detections with the ability to process 100,000 faces per second and it recommends the minimum image size to be 640x480 pixels .

VeriLook Test Result

Software works best with high resolution photos.

False Acceptance Rating set for 100%

All images matched 100% against the same image in the database. The score of 180 is interpreted as an exact match.

Free demo version used

VeriLook Aging Result

The results show that the photo from 1969 matched a photo from 2008 with a similarity score of 18 or 10%.

This result is comparable with the FaceSDK age identification test, where the same image from 1969 matched the same photo from 2008 with a 61.9% similarity rate.

Software Comparison TestLuxand FaceSDK - VeriLook

Tests were run using both Luxand FaceSDK 1.7 and VeriLook using the four photos seen here.

Similarity ratings varied from one software to the other.

Luxand FaceSDK results provided more results based on Similarity Rating than VeriLook .

VeriLook Identification and Authentication ResultsVeriLook Identification and Authentication Results

FaceSDK Identification and Authentication ResultsFaceSDK Identification and Authentication Results

Luxand – VeriLook Comparison

Conclusion

Luxand FaceSDK 1.7 works very well in identifying face similarity among people in a

group worked relatively well matching an image of the subject as a child VeriLook 3.2 had more limitations than FaceSDK, it only accepted high-res

images. results for the similarity test were lower than the FaceSDK

software PDA Security

none of the software tested was suitable for PDA security use

Further Work we recommend further work using 3-D face biometrics software

and scanners to find optimal solution for PDA security