Information Technology Reference
In-Depth Information
Age and Gender Estimation Using Shifting
and Re-scaling of Local Regions
Nawwar Ali, Chi-Fu Lin, Yuh-Shen Hsiung, Yun-Che Tsai, and Chiou-Shann Fuh
Department of Computer Science and Information Engineering,
National Taiwan University, Taipei 10617, Taiwan (R.O.C)
{nawwar.ali.j,daky1983,bear8039,jpm9ie8c}@gmail.com,
fuh@csie.ntu.edu.tw
Abstract.
A method for estimating age and gender using multiple local patches
is proposed in this thesis. We use the histogram of rotation-invariant local bi-
nary pattern as our features to train the SVM model. We further introduce the
shifting and scaling of the local patches to enhance the accuracy of the estima-
tion. Our proposed method not only provides accurate results but also can be
incorporated with other methods to further improve their accuracy.
Keywords:
Age and Gender Estimation, Local Invariant Local binary patterns,
Support Vector Regression, Haar Cascades, integral image, Supervised Descent
Method.
1
Introduction
Age and gender estimation is an important research topic in computer vision today; it
could be hard for humans to estimate the age and gender correctly, and automating
the process could offer many advantages. Age and gender estimation is to label a face
image automatically with the correct gender and the exact age or the age group of the
individual face.
Age estimation in humans is an ability that is developed early in life and it can be
fairly accurate. However it is usually more accurate in the estimator's own kind, i.e. if
you try to estimate somebody's age who is in your own age group and race and shares
with you the same gender you achieve better accuracy than other people. Humans
could use some other factors other than the face to estimate the age, such as the over-
all posture, hair color, and facial hair.
Some of the applications for such system would be to limit access for services to
users of a certain age; such as the sale of alcohol or tobacco or access to certain web-
site, or it could be used to retrieve photos of yourself when you were a certain age. It
could be used to allow different Graphical User Interface (GUI) to different users, if
the user appears to be of an older age the interface would use bigger size text.
Some of the challenges for accurate age detection are that aging is uncontrollable
non-linear process and that different people seems to age at different speeds. Another
problem is that the available databases for labeled facial images leave much be
