Gender Classification on Real-Life Faces - Advanced Concepts for Intelligent Vision Systems - page 322

Gender Classification on Real-Life Faces

Caifeng Shan

Philips Research
High-Tech Campus 36, 5656AE Eindhoven, The Netherlands
[e-mail]

Abstract. Gender recognition is one of fundamental tasks of face im-
age analysis. Most of the existing studies have focused on face images
acquired under controlled conditions. However, real-world applications
require gender classification on real-life faces, which is much more chal-
lenging due to significant appearance variations in unconstrained sce-
narios. In this paper, we investigate gender recognition on real-life faces
using the recently built database, the Labeled Faces in the Wild (LFW).
Local Binary Patterns (LBP) is employed to describe faces, and Ad-
aboost is used to select the discriminative LBP features. We obtain the
performance of 94.44% by applying Support Vector Machine (SVM) with
the boosted LBP features. The public database used in this study makes
future benchmark and evaluation possible.

Keywords: Gender Classification, Local Binary Patterns, AdaBoost,
Support Vector Machines.

1 Introduction

Gender classification is a fundamental task for human beings, as many social
functions critically depend on the correct gender perception. Automatic gender
recognition has many potential applications, for example, shopping statistics
for marketing, intelligent user interface, visual surveillance, etc. Human faces
provide important visual information for gender perception. Gender classification
from face images has received much research interest in last two decades.
In the early 1990s various neural network techniques were employed to rec-
ognize gender from frontal faces [1,2], for example, Golomb et al. [1] trained a
fully connected two-layer neural network, SEXNET, which achieves the recog-
nition accuracy of 91.9% on 90 face images. Recent years have witnessed many
advances (e.g., [3,4]); we summarize recent studies in Table 1. Moghaddam and
Yang [5] used raw image pixels with nonlinear Support Vector Machines (SVMs)
for gender classification on thumbnail faces (12

21 pixels); their experiments on
the FERET database (1,755 faces) demonstrated SVMs are superior to other
classifiers, achieving the accuracy of 96.6%. In [6], local region matching and
holistic features were exploited with Linear Discriminant Analysis (LDA) and
SVM for gender recognition. On the 12,964 frontal faces from multiple databases

×

Supported by the Visual Context Modelling (ViCoMo) project.