■ Fusion of Visible and Thermal Infrared Images for Robust
Face Recognition
■ Hyperspectral Image Analysis
for Biomedical Diagnostics and Target Detection
■ Signal
Restoration in Terahertz Imaging and Spectroscopy
■ Multisensor Image Registration
and Fusion
■ Evolvable Block-based Neural
Networks for ECG Signal Monitoring
Fusion of Visible and Thermal Infrared
Images for Robust Face Recognition
This work finds an
adaptive data fusion technique of visible and thermal infrared (IR) images for robust face recognition
regardless of illumination variations, partially supported by the Office of
Naval Research. The combined use of visible and
thermal IR image sensors offers a viable means for improving the performance of face recognition techniques based on single imaging
modalities. Visual imaging demonstrates difficulty in recognizing the faces in
low illumination conditions. Thermal IR sensor measures energy radiation from
the object, which is less sensitive to illumination changes and operable in
darkness. Data fusion of visible and thermal images can reproduce face images
robust to illumination variations. However, thermal face images with eyeglasses
may fail to provide useful information around the eyes since glass blocks a
large portion of thermal energy. Adaptive
data
fusion detects the presence of eyeglasses to enhance the quality of
visual-thermal image fusion in terms of information content for robust face
recognition.
Selected Publications:
■ J. Heo, S. G. Kong, B. Abidi, and M. Abidi,
“Fusion of Visual and Thermal Signatures with Eyeglass Removal for Robust
Face Recognition,” Proc. Workshop on
Object Tracking and Classification Beyond the Visible Spectrum (OTCBVS’04),
■ S. G. Kong, J. Heo,
B. R. Abidi, J. Paik, and M. A. Abidi, “Recent Advances in Visual
and Infrared Face Recognition - A Review,” Computer
Vision and Image Understanding, Vol. 97, No. 1, pp.103-135, January 2005. (Most
Cited Paper Award)
■ S. Moon and S. G. Kong, “Adaptive Fusion of Visible and Thermal Images
based on Multiscale Analysis for Face Recognition,” Proc. IEEE Int’l Conf.
on Computational Intelligence for Homeland Security and Personal Safety
(CIHSPS’06), Alexandria, VA, Oct. 2006.
■ S. G. Kong, J. Heo, F. Boughorbel, Y. Zheng, B.
R. Abidi, A. Koschan, M. Yi, and M. A. Abidi, “Multiscale
Fusion of Visible and Thermal IR Images for
Illumination-Invariant Face Recognition,” International Journal of Computer Vision, Vol. 71, No. 2, pp.215-233, Feb. 2007.
Hyperspectral Image Analysis for Biomedical Diagnostics
and Target Detection
This
research aims at developing hyperspectral imaging and signature classification
techniques for real-time, non-invasive diagnosis of skin cancers. The
application focus involves the study and correlation of
reflectance/fluorescence image signals with neoplastic properties of normal and
tumor tissues. A 
technical
innovation of this project is the combination of an advanced computing
algorithm based on the support vector machine with a real-time hyperspectral
imaging system for detecting small differences in reflectance and fluorescence
profiles of normal and malignant tumor tissues. This approach leads to
significant advances in effective and rapid detection of tumors over large areas
of organs and in the understanding of cancer in general. The developed skin
cancer detection technique will be translatable to other diseases in other
organ sites as well changing the future of diagnostic medicine. Early phase of
this research has been successfully applied to the detection of skin tumors on
poultry carcasses for food safety inspection.
Selected
Publications:
■
S.
G. Kong, Y. R. Chen, I. Kim, and M. S. Kim, “Analysis of
Hyperspectral Fluorescence Images for Poultry Skin Tumor Inspection,” Applied
Optics, Vol. 43, No. 4, pp.824-833, February 2004.
■
I.
Kim, Y. R. Chen, M. S. Kim, and S. G. Kong, “Detection of Skin
Tumors on Chicken Carcasses using Hyperspectral Fluorescence Imaging,” Transactions
of the American Society of Agricultural Engineers, Vol. 47, No. 5,
pp.1785-1792, 2004. (ASABE Honorable Mention Paper
Award)
■
S.
G. Kong, Z. Du, M. Martin, and T. Vo-Dinh, “Hyperspectral
Fluorescence Image Analysis for Use in Medical Analysis,” Proc. of SPIE
Conf. on Biomedical Optics,
■
S.
G. Kong, M. Martin, and T. Vo-Dinh, “Hyperspectral
Fluorescence Imaging for Mouse Skin Tumor Detection,” ETRI Journal,
Vol. 28, No. 6, pp.770-776, December 2006.
■
Z.
Du, M. K. Jeong, and S. G. Kong, “Band Selection of
Hyperspectral Images for Automatic Detection of Poultry Skin Tumors,” IEEE
Transactions on Automation Science and Engineering, Vol. 4, No. 3,
pp.332-339, 2007.
Signal Restoration in Terahertz Imaging and Spectroscopy
The part of the
electromagnetic spectrum (0.1?0 THz) between microwaves and far infrared refers
to Terahertz radiation. Terahertz beams easily penetrate dielectric substances
such as paper, plastics and ceramics, which are opaque at optical frequencies
and provide very low contrast for X-rays. These materials are relatively
non-absorbing in this frequency range, yet different materials may be easily
discriminated on the basis of their refractive index, which is extracted from
the THz phase information. Unlike X-rays, Terahertz beams can be focused and
are capable of producing phase-sensitive spectroscopic images with signature capability.
The spectral information of the transmitted and/or scattered radiation could be
used to detect hazardous chemicals, threat objects in luggage or concealed
weapons, and defects in semiconductor wafers and packages. Low photon energy
level of Terahertz spectrum is particularly attractive for the imaging of
biological tissues with no harmful ionizing radiation. THz source can be used
for high-resolution remote subsurface imaging, with spatial and depth
resolution better than 1 mm, enabling differentiation of skin or breast cancers
from normal tissues and tooth cavities. This research covers enhancement of THz
waveforms and denoising, spectral signal classification, high-resolution image
reconstruction based on the fusion of peak intensity and phase information, and
segmentation techniques for the visualization and classification of terahertz
images.
Selected
Publications:
■
S.
G. Kong and D. H. Wu, "Signal Restoration
from Atmospheric Degradation in Terahertz Spectroscopy," Journal of
Applied Physics, Vol. 103, No. 11, 113105 (6 pages), June 2008.
■
S.
G. Kong and D. H. Wu, “Terahertz Time-Domain Spectroscopy for Explosive Trace
Detection,” Proc. IEEE Int’l Conf. on Computational
Intelligence for Homeland Security and Personal Safety (CIHSPS’06),
■
S.
G. Kong and D. H. Wu, “Signal Restoration from Atmospheric Degradation in
terahertz Spectroscopy,” Journal of Applied Physics, in review, 2008.
Multisensor Image Registration and Fusion
The goal of this
research is to develop a multisensor image registration technique based on the
combination of mutual information computed from intensity and gradient
magnitude of an image pair. Image registration refers to the process of
aligning multiple images of the same scene taken by different imaging sensors,
at different times, and/or from different viewpoint. Multiple images obtained
from different types of imaging sensors (e.g. anatomical (CT) and functional
(PET) medical images or radar and optical target images) may contain complementary
information on the object of interest. Multisensor image data must be
integrated to reveal the information useful for medical diagnosis or target
identification. Maximizing the mutual information has been widely used as an
effective method in various image registration applications. However,
individual mutual information using either image intensity or gradient
magnitude often demonstrates limitations in the registration of multisensor
images. The image registration technique proposed in this effort synergistically
combines the mutual information of intensity and gradient magnitude as robust
measure of multisensor image registration.
Selected
Publications:
■
O.
Kwon and S. G. Kong, “Multimodality Image Registration using Combined Mutual
Information of Intensity and Gradient,” World
Congress on Medical Physics and Biomedical Engineering,
■
W.
Jiang and S. G. Kong, “Combined Mutual Information
of Intensity and Gradient for Multisensor Image Registration,” in preparation, 2008.
Evolvable Block-based Neural Networks for ECG Signal Monitoring
This work aims at
developing evolvable neural networks that reconfigure their structures and
connection weights autonomously in dynamic operating environments. The
block-based neural network model consist of a 2-D array of basic blocks that
can modify network structure and connection weights using evolutionary
algorithms to be implemented on reconfigurable digital hardware. Block-based
neural networks demonstrated the potential for analyzing electrocardiogram
(ECG) signals to monitor human health conditions online that are insensitive to
variations over individuals, time of day, and under different body conditions.
People working in dangerous environments (e.g. military personnel, firemen, and
truck drivers) as well as older people will benefit from constant monitoring of
their health conditions for prediction of various dangerous states such as
detection of losing consciousness and heart infarct.
Selected
Publications:
■ S. W. Moon and S.
G. Kong, “Block-based Neural
Networks,” IEEE Transactions on
Neural Networks, Vol. 12, No. 2, pp.307-317, March 2001.
■ W. Jiang, S. G. Kong, and G. Peterson, “ECG Signal Classification using Block-based
Neural Networks,” Proc.
International Joint Conf. on Neural Networks,
■ S. Merchant, G. D. Peterson, S. Park, and S. G.
Kong, “FPGA Implementation of Evolvable Block-based Neural Networks,” Proc. Congress on Evolutionary Computation,
Vancouver, July 2006.
■ W. Jiang and S. G. Kong, “A Least-Squares Learning for Block-based Neural
Networks,” Dynamics of Continuous, Discrete and
Impulsive Systems, Vol. 14, No. S1, pp.242-247, 2007.
■ W. Jiang and S. G. Kong, “Block-based Neural Networks for Personalized ECG Signal Classification,” IEEE Transactions on Neural Networks, Vol. 18, No. 6, pp.1750-1761,
November 2007.