Age and Gender Estimation Using Shifting and Re-scaling of Local Regions - Technologies and Applications of Artificial Intelligence - page 11

of 60 Fig. 8. The accuracy of age cla

pixels

assification from the face region (single patch) with shifting o

6 Conclusion and Future Work

Age and gender estimation
aging and the subtle diffe
age and gender estimation
We further demonstrated th
further increase the accurac
During our experiments,
overlapping helped increase
of the face (instead of just
with the estimation of age
further modified to include

n is a challenging problem due to the non-linearity of
erence between individuals. In this thesis we propose
algorithm based on the fusion of multiple local patch
he effectiveness of shifting and scaling the local patche
cy of the performance.
, we find out that increasing the size of the shifting beyo
es the accuracy. This helps to demonstrate that all patc
t the main facial features) hold information that can h
and gender. Using this information the algorithm could
other patches to increase the accuracy.

the
e an
hes.
s to

ond
ches
help
d be

Acknowledgement. This r
Technology of Taiwan, R.O
103-2221-E-002-188, and
Digital Optics, PolarLink, a

research was supported by the Ministry of Science
O.C., under Grants NSC 102-2221-E-002 -177 and MO
by Winstar Technology, Test Research, Qisda, Lum
and Lite-on.

and
OST
mens

References

1. Boser, B.E., Guyon, I., V
In: Proceedings of the An
PA, pp. 144-152 (1992)

Vapnik, V.: A Training Algorithm for Optimal Margin Classifi
nnual Workshop on Computational Learning Theory, Pittsbu

iers.
urgh,