Human visual system based facial recognition for autonomous systems and applications.
thesis presents a real-time facial recognition system utilizing our human visual system
algorithms coupled with classical Logical Binary Pattern (LBP) feature descriptors.
Specifically, we first use the Weber's Law based human visual system to obtain
decomposed images. Then, features are extracted using logarithmic Logical Binary
Patterns (LBP). The contributions of this work i... read morenclude introducing region weighted
models for facial components. We investigate two models, Hybrid region weighted model
and Hybrid-Holistic region weighted model, and compare and contrast the performance on
public databases of faces. Finally, the similarity ranking is obtained by fusing the
chi-squared distance evaluated from each individual facial region. The system can
quickly find and rank the closest matches of a test image to a database of stored
images. For our prototype application, we supplied the system testing images and found
their best matches in the database of training images. This system can also be applied
in many real life applications. One of them is automatically matching composite sketches
to facial photographs. Different sketches hand drawn by artists or composite sketches
synthesized using facial composite software can be compared to a database of photographs
for the closest match. This is a particularly useful application for law enforcement
agencies. Furthermore, other applications that could utilize this work include finding
missing children or victims of human trafficking. Our methodology produces a low cost
and efficient detection and recognition that weights more important facial features that
may still be important to identify individuals, even though many years may have passed
between the times the last known photograph was taken. Often times, age progressed
composite sketches are created to aid in the search. These sketches can be used in our
system to help find individuals that have gone missing or to help locate wanted
criminals. Finally, we investigate the limitations of the system by applying the
algorithms for other non-human facial features, namely matching animals. These
experiments reveal the specialized features that researchers must take into
consideration when searching for any subject possessing a "face". Experimental results
have clearly shown the promising performance and a great value to law enforcement
Thesis (M.S.)--Tufts University, 2015.
Submitted to the Dept. of Electrical Engineering.
Advisor: Karen Panetta.
Committee: Sampathkumar Veeraraghavan, and Ronald Lasser.
Keyword: Electrical engineering.read less