Open Journal of Applied Sciences, 2013, 3, 53-57
Published Online March 2013 (http://www.scirp.org/journal/ojapps)
Copyright © 2013 SciRes. OJAppS
Multi-Faces Image Recognition
Jung-soo Han, Jeong-heon Lee, Gui-Jung Kim
Division of Communication & Information, Baekseok University, Cheonan, Korea
National Information Society Agency, Seoul, Korea
Department of Biomedical Engineering, Konyang University, Nonsan, Korea
Email: jshan@bu.ac.kr, opendori@gmail.com, gjkim@konyang.ac.kr
Received 2012
ABSTRACT
The paper suggests a great improvement in the recognition rate and speed by overcoming the disadvantage of 3D image
processing taking a long time as well as that of 2D images weak in postures through correcting shooting angles with 3D
data taken by multi cameras followed by application of the developed hybrid face recognition technology.
Keywords: Hybrid; Face Recognition; 3D; 2D; Image Recognition
1. Introdu ction
As of 2009, the ISO's two groups, SC29/WG11 (MPEG)
and SC37, have been standardized in terms of face
detection and recognition. Although technology of face
detection and recognition has achieved substantial results,
research on face detection and recognition technology
related to an environment integrated into CCTV has just
started. In Korea, as CCTV-based face detection and
recognition technology serves a crucial role in preventing
terror and crime and arresting criminals, its necessity is
consi dered at national level. Currently, when it comes to
arresting criminals, not only CCTVs installed in the
nation cannot identify personal characteristics, but also a
read rate is low, caused by light effects such as slow face
recognition speed, lighting, and diverse facial pose and
expression, although the face is detected or recognized[1].
As such, to overcome limited technology on
CCTV-based face detection and recognition technology,
research is conducted on detection and recognition tech-
nology on various faces such as a recent IR-based face
recognition technology strong in lighting. Particularly,
developing outstanding face detection technology is a
prerequisite for development of high-performance facial
recognition system. However, the existing face detection
method is to find a face location in an image, which has
many problems to solve including face judgment, differ-
ent-sized face search, exact-domain face search, face de-
tection speed, result combination, along with problems in
detecting face. A CCTV stands for Closed Circuit Televi-
sion, communication equipment that enables a specific
person to receive images collected by shooting equipment
installed in a place after transmission through closed
wired/wireless channels. Currently, although security
thr ea t s occurs to CCTVs, regarding CCTV installment
and operation, the law of it is not properly established,
and the system of it is not properly conducted [2-5].
In this paper, we explained unmanned security control
system and simulated faces recognition process .
2. Unmanned security Robots
Unmanned security surveillance robots can recognize
authori zed/una uthor ized by real-time processing images
input by HD-level camera, save a relevant image to
transmit via control server, store transmitted image ac-
cording to a policy, which enables a controller to provide
a preview function after searching saved images through a
search function. Unmanned security surveillance robot's
image recognition system has the following composition
elements.
Such as Figure 1, it serves as a function to detect a face
in a HD-level image; extract and record face characteristic
data to compare with authorizer. With regular hour (day-
time) fixed mode, it serves as an alert surveillance
(HD-level CCTV function), and automatically changes
into move mode in set time (e.g. night time, after school,
holidays, etc.), alert surveillance is performed by moving
in a set place and a surveillance poor site. It can be re-
charged in a fixed mode, and saved images can be trans-
mitted to the control center in a move mode when an alert
surveillance. In a move mode, when an unauthor i zed is
found, a warning message is transmitted to the unautho-
rized by speaker warning and lamp alarm. Saved images
during an alert surveillance are saved in the inside of the
robot, and when it is recharged, they are transmitted to the
control center to save. To compare information between
authori zed/unaut horiz ed, a captured frame (or image) in a
J. HAN ET AL.
Copyright © 2013 SciRes. OJAppS
video clip should be solely extracted and transmitted to
the center to be compare with DB. In case of lights-out in
night time alert, detect a moving object by sensing with a
built-in moving detecting sensor, and when detecting
something, a lamp of the robot is lighted on to save a re-
levant image. Also it serves as a role to upload images
saved in robot to NAS storage through FTP. Robot Agent
receives IP, ID, and PASS from NAS storage and uploads
a file by creating a directory saved at NAS server through
a regulation of saving file name [6, 8].
Figure 1. Unmanned Security Robot System Struc-
ture
3. Image Recognition System
3.1. Human and Face Recognition
To develop a technology employing algorithm of
Video Capturing Module that obtains real-time im-
ages through network cameras and Face-Detecting
Modul e that detects facial part from input images, and
Face Recognition Module that processes the com-
parison and recognition between extracted images and
database, unmanned monitoring robot platform and face
reco gnitio n-based identification technology are necessary.
Un mann ed security monitoring robot in the form of a
differential drive mobile robot has each sub-controller for
an actuator and control panel and each controller is or-
ganically connected to main controller to perform com-
mand. The battery for the motor to run should be re-
charged in 2 hours and last more than 4 hours for moving,
operating video and communication equipment. In addi-
tion, it is designed to run on a road. Moving up or down
stairs and on uneven terrain are not placed in design con-
sideration. It is also designed to change speed, consider-
ing running circumstances at average travelling speed of
about 4km/h and maximum speed of more or less 6km/h
and speed change is made in two modes: automatic and
operation by controller [7, 9].
The face-recognition system is a system that detects
face image from video data, extract distinctive data the
extracted face image and process matching work to iden-
tify a person. To do the task, the system consists of face
detection module, distinction extraction module and
matching engine that is a recognition module. RTSP ana-
lyzes images obtained and works on identifying process
based on face. This consists of face detection module and
distinction extraction module. We can find many re-
searches related to the methods for face detection and
recognition, such as Adaboost, Gabor 2D Wavelet com-
bination, stochastic color model and deformable template.
But because some of their core parameters used in algo-
rithm are fixed, they have shortcomings that they aren’t
flexible under changing circumstances. The unmanned
security monitoring robot of this study needs multiple
face detection and recognition and especially the algo-
rithm suitable for irregular images input due to moving.
In the process of considering many methods of face
recognition and testing their performance, this study used
Gabor-kernel wavelet transformer, which is effective for
such changes as illumination, facial expression or pers-
pective projection in order to extract facial distinction
data, and applied principal component analysis based on
dual non-linear mapping technique that realizes the effec-
tive transformation of distinction data for the purpose of
face recognition. This method placed in consideration not
only the statistical dispersion of Gabor distinction data but
also spatial information of human face. Accordingly these
distinction date converted after non-linear mapping had
high discrimination and as a result, these data can be in-
sensitive to the changes in various conditions such as il-
lumination, facial expression or perspective projection.
Developing face-based identification technology re-
quires real-time multiple face detections and recognition,
which are not vulnerable to lightings and posing. Face
recognition system can be divided largely into distinc-
tion-based and appearance-based method. Distinc-
tion-based face recognition system is a method to use the
geometrical information of a face or distinctive compo-
nents of a face such as eyes, nose, lips and chin in order to
obtain information about their relationship or the com-
bined shapes of these components. This technique is fast
in processing time, simple in structure and easy to recog-
nize face. However, it has limitation that it couldn’t detect
the distinctive components of a face depending on the
angle a face tilts, so that it is very vulnerable to lighting
and poses. Therefore, many ways have been suggested get
over the limits. Appearance-based face system is one
most used in the field of face recognition. It uses a learned
model by the collection of learning images to recognize
face. This system use many facial recognition techniques
such as Eigenface produced by Principal Component
Analysis (PCA), Fisherface created by Linear Discrimi-
nate Analysis (LDA), Neural network (NN), and Support
Vector Machine(SVM) [10].
54
J. HAN ET AL.
Copyright © 2013 SciRes. OJAppS
3.2. Intellectual Image Monitoring System
Intellectual Image Monitoring/Controlling System con-
sists of robot-operation management server and im-
age-recognition management server. This system receives
image information in real time from network camera built
in a security monitoring robot, displays concerned infor-
mation on 9 screens, and monitors and remote control the
status of a robot platform. Also the system includes inte-
grated management DB that controls the storage and dis-
posal of transmitted images and data necessary for robot
operation and system controlling, image recognition DB
that manages data of image recognition and licens-
er/ non -licenser, and the function of control and manage
the entire system in interlocking with other systems. HD
Network Camera Control Module use a camera of HD
quality to receive images in real time through robot and
supports and control the camera remote to make it (robot)
act actively against unexpected events [10, 11].
As DB operating system that receives, stores, deletes or
search images RTSP recorder saves in accordance with
related policy, Image Storage DB performs a role of
managing integrated controlling server. Im-
age-recognition management server saves facial informa-
tion (facial distinction date and concerned facial thumb-
nails) from a robot’s Image Analyzer to set up image
tracking and face recognition DB in storage. After saving
them, it compares the new information with the data al-
ready registered in DB to discriminate and then send the
results back to the robot.
3.3. Security Monitoring Service
Escaping from the conventional mode of one-way securi-
ty monitoring system and focusing on mobility and
self-controlling, which are characteristics of a robot, this
paper places its objective in the development and com-
mercialization of intellectual security monitoring/ control-
ling system that can run unmanned 24 hours through
grafting with HD high quality image-based intellectual
image technology of a network camera.
To provide efficient security monitoring service in op-
eration of 24-hour image security monitoring system, a
monitoring robot is designed to perform video-securing
work in a gate (at fixed position) during daytime and at
night it moves automatically on specified routes (set on a
map) to monitor any intruder and actively cope with un-
expected situations. <Table 1> shows scenarios by situa-
tion for the operation of unmanned security monitoring
robot [1, 4].
3.4. Face Recognition System Process
Face Recognition System is a system that extracts face
from input images, and recognizes who is whom through
similarity assessment process and characteristic data reg-
istered by extracting characteristic data for recognition in
extracted face. To achieve this, it is composed of face
detection module, characteristic extraction module and
similarity assessment engine, a recognition module. An
assessment module for face similarity is used as a
PCA-specific extractor input based on Gabor-kernel for
facial range image extracted through Adaboost learning
algorithm. A face recognition system calculates an Eucli-
dean distance between authorizer and unauthorized r e gis-
tered and factors of characteristic data extracted, which
maps face in the closest distance. It is a module that cal-
culates similarity between registered faces, recognizing
whom is out of authorizers through face recognition, and
it serves as a function of grouping as unauthorized, oth-
erwise[12, 13].
4. Faces Recognition Process
Face Recognition System is a system that extracts face
from input images, and recognizes who is whom through
similarity assessment process and characteristic data reg-
istered by extracting characteristic data for recognition in
extracted face. To achieve this, it is composed of face
detection module, characteristic extraction module and
similarity assessment engine, a recognition module. A
process of face recognition system is as follows.
Figure 2 shows face image registration process from
structured system. After 1 or 2 people registered, and then
if the face of CCTV images coming into the camera
compare with the images on the server and it matches the
data stored in the image a nd displayed the name recogni-
tion.
Figure 3 shows the register and at the same time, three
CCTV images showed on the recognition process. How-
ever, when you receive the generic self image is displayed
on the tail-light rod, grow up. Depending on the distance
of the camera or if the far is recognition, but the feature
extraction is unable, just when I came close with a cam-
era-aware image possible.
Figure 2. Face Image Registration
55
J. HAN ET AL.
Copyright © 2013 SciRes. OJAppS
(a) 3 faces recognition
(b) un-registration search
Figure 3. Faces Recognition Simulation
5. Conclusion
In conclusion, this paper was conducted for a method to
protect personal privacy at maximum at a CCTV
environment following the trend of international and
domestic standardization, and a framework modeling
method that analyzes face detection and recognition
system security widely applicable, which finds an
implementation method to develop a standard model that
can be demonstrated on important service (client ID
management, criminal arrest, terror prevention and
access management, etc.) based on information
protection and biometric reference standard development,
and delineates standardized plans for personal
information protection on requirements for related bio
information and reference management server,
mechanism, secured system of database, and
management system, image surveillance system and
CCTV-based image surveillance system.
In this paper, methods are suggested for developing a
basic algorithm and framework for 2D-based face
recognition where face images for reference DB for face
recognition are collected, converted to data as a reference
for similarity using an engine for extracting
characteristics, with the fabricated face image
information mapped with DB as well as a module for
image input/output interface, image pre-treatment,
normalization and characteristics extraction. With
development of a technology to map the fabricated face
image information with user DB and measure similarity,
a technology that can overcome the limitations in multi
processing ability of the existing fingerprint recognition
system and save the maintenance cost of RFID card
method has been suggested.
To achieve this, by manufacturing an unmanned
security surveillance robot, images input through
HD-level camera are processed real-time to identify
author i zed/u na ut ho r i ze d , save a relevant image to
transmit to the control server, and keep transmitted
recorded images according to a policy, which offers a
person in charge to preview them after saved image
search through search function. It is a module to upload
images real-time to the integrated control system by
saving transmitted module and image real-time that
enables monitoring at the integrated control system
application. The face recognition system is a system that
detects face from input images, and recognizes who is
whom through similarity assessment process between
characteristic data registered by extracting characteristic
data to recognize extracted face.
6. Acknowledgements
This work was supported by the Security Engineering
Research Center, granted by the Korea Ministry of
Knowledge Economy.
REFERENCES
[1] Frederic Dufaux, Scrambling for Privacy Protection in
Video Surveillance Systems, IEEE TRANSACTIONS
ON CIRCUITS AND SYSTEMS FOR VIDEO TECH-
NOLOGY, 2008, 18(8).
[2] Kenichi YABUTA, Privacy Protection by Masking
Moving Objects for Security Cameras, IIEICE Trans-
actions on Fundamentals of Electronics, Communications
and Computer Sciences, Volume E92.A, Issue 3, 2009 :
919-927.
[3] Mitsuji Muneyasu, An Implementation of Privacy
Protection for a Surveillance Camera Using ROI Coding
of JPEG2000 with Face Detection, IIEICE Transac-
tions on Fundamentals of Electronics, Communications
and Computer Sciences, 2009, E92-A(11) : 2858-2861.
[4] E. Yu and J. Aggarwal, "Human action recognition with
extremities as semantic posture representation," Proc. Int.
Conf. Computer Vision, Pattern Recognition, 2009: 1-8.
[5] TI, TMS320C64x+ DSP Image/Video Processing Li-
brary Programmer's Guide, Texas Instruments, Dallas,
Texas, May, 2008.
[6] TI, Vision Library Application Programming Interface
Reference Guide, Texas Instruments, Dallas, Texas,
Nov., 2009.
[7] Chris Stauffer and W.E.L Grimson, Adaptive Back-
ground Mixture Models for Real-time Tracking,
Proc.IEEE Conference on Computer Vision & Pattern
Recognition, 1999, 2 : 246- 252.
[8] A Al-Habaibeh, F Shi, N Brown, D Kerr, M Jackson and
R M Parkin, "A novel approach for quality control system
using sensor fusion of infrared and visual image
processing for laser sealing of food containers," MEA-
SUREMENT SCIENCE AND TECHNOLOGY, 2004 :
1995~2000.
[9] Philippe Simard, Norah K. Link, Ronald V. Kruk, "Eval-
uation of Algorithms for Fusing Infrared and Synthetic
Imagery," Enhanced and synthetic vision 2000. Confe-
rence, Orlando FL, 2000, 4023 : 127-138.
[10] C. Wren, A. Azerbayejani, T. Darrel, and A. Pentland,
Pfinder: real-time tracking of the human body. IEEE
Trans. Pattern Analysis and Machine Intelligence, 1997,
19(7): 780-785.
[11] S. Kang, J. Paik, A. Koschan, B. Abidi, and A. Abidi,
Real-time video tracking using PTZ cameras. Proc.
56
J. HAN ET AL.
Copyright © 2013 SciRes. OJAppS
SPIE 6th International Conference on Quality Control by
Artificial Vision, 2003, 5132 : 103-111.
[12] P. Azzari, L. Stefano, and A. Bevilacqua, An effective
real-time mosaicing algorithm apt to detect motion
through background subtraction using a PTZ camera.
IEEE Conf. Advanced Video and Signal-Based Surveil-
lance, 2005 : 511-516.
[13] G. A. Ramirez and O. Fuentes, Multi-pose face detec-
tion with asymmetric HAAR features. Applications of
Computer Vision, 2008. WACV 2008. IEEE Workshop,
2008 : 1-6.
57