Existing System for Face Recognition

(A) Facial Recognition:-

In the last twenty years, the computer-based facial recognition field has expanded rapidly.
Several algorithms have been introduced and improved to the point where computers can rival
humans in accuracy of facial recognition [14]. In order to develop our product, we need to
understand how we identify faces, and to understand and evaluate the different existing facial
recognition algorithms and examine existing applications of this technology. Sinha, Balas,
Ostrovsky, and Russell outline nineteen basic results regarding human facial recognition,
including many of the methods that humans use to identify faces [6]. They show that the study of
human processes involved in facial recognition and the artificial algorithms being used for facial
recognition systems are inextricably linked together. The human brain can recognize faces in 120

milliseconds (ms). In order to achieve a useful system, the algorithm we choose must have near
realtime feedback. Several real time algorithms have been developed in recent years. Ross Bev-
8 eridge of the Colorado State University evaluated the efficiency and accuracy of the
algorithms, Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA), Elastic
Graph Matching (EGM), and Bayesian Intrapersonal/Extrapersonal Image Difference Classifier
(BIC) [15]. A study by Krishna, Little, Black, and Panchanathan also evaluated these algorithms
with respect to changes in illumination and pose. The LDA and PCA algorithms were found to be
superior. LDA was fastest while PCA was the most accurate [3]. We will be evaluating the four
algorithms (PCA, LDA, EGM, and BIC) to determine what works best for our purposes, as our
conditions may be slightly different than those of previous studies. Another algorithm recently
developed is Luxands FaceSDK, a facial recognition system that supports recognition for still
images and real time video streams. It is mostly used in the entertainment industry to create real
time animations. As outlined by Zhao, Chellappa, Phillips, and Rosenfeld [16], these algorithms
may be susceptible to several well known challenges including pose, illumination, and
resolution. However, over the past several years, major improvements have been made to these
baseline algorithms. In an experiment by Alice J. OToole, seven facial recognition algorithms
were compared with humans on face matching tasks. Out of the seven algorithms, three were
better at recognizing faces than humans were [14]. Though illumination still presented problems,
OTooles study shows that current algorithm capabilities compete favorably with human ability
to recognize faces. Another challenge to consider is the movement of visually impaired subjects,
which may cause the acquired faces to be blurred. We need to develop image deblurring tools
and feed the restored face images to the recognition algorithms. 9 Systems for recognizing faces
do currently exist. However, many of the existing facial recognition systems are created for
security rather than for the visually impaired [17]. Even so, these systems show that it is possible
for a recognition system to recognize faces acquired under controlled conditions at a recognition
rate of 99.2% in near realtime. One system, developed by Krishna et al., uses a PCA algorithm
and was designed with the visually impaired in mind. Krishnas system does not use stereo
cameras, nor was it tested on visually impaired users. Krishnas system also did not have a facial
expression recognition feature. Our system will be designedwith the advice of the visually
impaired to provide a facial recognition system that would benefit them the most.

iCare Interaction Assistant

Krishna et al. [3] have developed the iCare Interaction Assistant, an assistive system that

acquires video from a pinhole aperture analog CCD camera embedded in a pair of eyeglasses,

digitizes it and then transmits it over a USB cable to a tablet PC. The video is analyzed

to detect faces using adaptive boosting which are passed to a face recognition module that

utilizes the Principal Components Analysis (PCA) and LinearDiscriminant Analysis (LDA)

algorithms. If a face is recognized in 5 consecutive frames, the name of the identified individual

is converted from text to speech and transmitted to the user via head phones. One main concern

expressed by Krishna et al. is that even though some publicly available face databases contain

images captured under a range of poses and illumination angles, however, none of them use a

precisely calibrated mechanism for acquiring these images, nor is each image explicitly

annotated with this information. Krishna et al. have therefore, put together their own database

called FacePix [29] which contains face images of 30 people with pose angles and illumination

angles between -90 and +90 degrees annotated in 1-degree increments. These two methods were,
therefore, selected for the face recognition module of the system. The system was tested with 10

known individuals and PCAs performance was found to be better than (or similar to) LDA.

Since PCAs computational complexity is also lower than that of LDA, hence it is the preferred

algorithm for future