Face expression recognition using Scaled-conjugate gradient Back-Propagation ...
LinkedIn_Arash_Lab
1. Kyle Mabry
January 2017
Seventy percent of police departments in the United States utilize license plate scanners
to identify stolen vehicles, drivers with suspended licenses, and hotspots for crime (Hsu et al.
2014). What if these scanners could be advanced further by enabling them to identify an
individual simply by his or her unique voluntary eye-pursuit patterns? This type of system would
require the construction of a “visual fingerprint” (VF) as well as a “visual fingerprint machine”
(VFM) that can accurately identify VFs. In order to develop a VFM we must first define the
visible and recordable components of eye movements.
A functioning visual system is able to readily engage in voluntary eye movements
(VEM). VEM is the ability to execute eye movements within the visual field, with the observer’s
control, and is comprised of two sub-types: smooth pursuit (SP) and saccades (SCS) (Grönqvist
et al. 2005). SP is responsible for tracking a moving target across the visual field at a low
velocity, i.e. a few visual degrees per second (Mann et al. 1997; Dodge et al. 1903), but SP is
often unable to follow the exact velocity of a moving target and subsequently falls behind
(Portron et al. 2017). When SP falls behind the moving target, SCS are responsible for
compensating by initiating a rapid jerk eye-movement to catch up to the target; the point at
which a saccade is initiated varies from individual to individual (Grönqvist et al. 2005).
Furthermore, marked differences exist between the effectiveness of VP in the horizontal axis vs.
the vertical axis. Generally, SP in the horizontal direction is smoother (i.e. less mixed with
saccades) than in the vertical direction (Rottach et al. 1997).
It is known that approximately 40-60% percent of the human brain is involved in visual
processing (Felleman et al. 1991). Every brain is structurally and functionally unique due to
varying individual experiences that help shape its evolution. Combining these two facts with our
knowledge of the components of vision, it is sensible to assume that each individual has a
uniquely designed VP tracking system or VF. It is the goal of this study to classify individual
VFs based on the unique components of vision, store the VFs, and later use them to identify the
individual via VP testing.
2. Works Cited
C A Mann, M J Morrow (1997) Effects of eye and head position on horizontal and vertical
smooth pursuit. Invest. Ophthalmol. Vis. Sci.;38(3):773-779.
DJ Felleman, DC Essen Van. (1991) Distributed hierarchical processing in the primate cerebral
cortex. Cereb Cortex. PMID: 1822724
Dodge, R. (1903). Five types of eye movement in the horizontal meridian plane of the field of
regard. American Journal of Physiologyj 8, 307-329.
Grönqvist, Helena; (2005) Developmental asymmetries between horizontal and vertical tracking.
Vision Research 46 (2006) 1754–1761
Hsu Jeremy; 70 Percent of U.S. Police Departments Use License Plate Readers IEEE Spectrum
2014 http://spectrum.ieee.org/cars-that-think/transportation/sensors/privacy-concerns-grow-as-
us-police-departments-turn-to-license-plate-readers
Portron Arthur (2017) Sustained smooth pursuit eye movements with eye-induced reverse-phi
motion. Journal of Vision January 2017, Vol.17, 5. doi:10.1167/17.1.5
Rottach, K. G., Zivotofsky, A. Z., Das, V. E., Averbuch-Heller, L., Disc-enna, A. O.,
Poonyathalang, A. (1997) Comparison of horizon- tal, vertical and diagonal smooth pursuit eye
movements in normal human subjects. Vision Research, 36, 189–2195.