You are on page 1of 1

‘Stereo-biometrical interface in real time’

My thesis work is devoted to ‘Stereo-biometrical interface in real time’. This


topic lies in the field of pattern recognition.
Face recognition is of great importance in many applications such as personal
identification, employee access to high security areas, human-machine interfaces.
The main advantage of face recognition as a biometric is its convenience and non-
invasiveness. Most of the research in face recognition has focused on 2D intensity
images. However, results from recent studies clearly show that the performance of
these traditional 2D face recognition approaches are adversely affected by varying
lighting conditions and especially by varying pose. The 3D facial surface
information is invariant under different lighting. The problem of varying pose can
also be corrected rather easily by rotating the 3D facial surface around the
symmetry lines.
My present investigation includes 2D and 3D methods of human biometric parameters
identification which are used for intuitive interface. Methods of human biometric
parameters identification include: algorithms for getting biometric parameters,
classification algorithms, system learning algorithms, decision making algorithms.
In order to do the face recognition I first of all need to capture the 3D facial
surface. There are three main ways for 3D facial surface capture. The first one is
based on structured lighting, in which a grid is projected on a face and the 3D
facial surface is calculated. The second is passive stereo using two cameras to
capture a facial image and using a computational matching method, the 3D facial
surface is created. Finally the third method is based on the use of laser range-
finding systems to capture the 3D facial surface.
After surface capture I need some feature detection and ratio computation
algorithms. Then I’m comparing the achieved values with the knowledge base and
deciding how to classify the surface.
Except person identification classification algorithms can also perform gender,
age and ethnicity classification although this can even be more difficult task.
But in some applications, for example, age distinction can be used for parental
control which is very useful nowadays.
We can use multimodal biometric system in order to improve the accuracy of the
system. Multimodal biometric system is the system that consists of multiple
sensors. These are generally more reliable due to the presence of multiple pieces
of evidence. There are various types of fusion that are possible when combining
multiple sensors: 1) fusion at image level; 2) fusion at feature level; 3) fusion
at scoring level; 4) fusion at decision level. Early stage fusion is known to have
better performance than later stage fusion.