Identifying Gender from Images of Faces

SHIVAM JHA AKSHYA CHAUDHARY

( CSJMA13001390160 ) ( CSJMA13001390157 )
Department of Information Technology, UIET CSJM Department of Information Technology, UIET CSJM
University Kanpur University Kanpur
-------------------------------------------------------------------------------------------------------------------------------------------------
ABSTRACT

The objective of this project is to identify the gender of a person by looking at his/her photograph. This is a case
of supervised learning where the algorithm is first trained on a set of female and male faces, and then used to
classify new data. We have not taken genders other than Male and Female into account. A preliminary algorithm
is run to make sure that an image is that of a human before classification begins.

Keywords: Machine Learning, Pattern Recognition, Gender Classification, Face Detection, SVM, Supervised Learning,
PCA, ICA, Fisherface.

I. Introduction

Previous research has shown that our brain has specialized nerve cells responding to specific local features of a
scene, such as lines, edges, angles or movement. Our visual cortex combines these scattered pieces of information
into useful patterns. Automatic face recognition aims to extract these meaningful pieces of information and put them
together into a useful representation in order to perform a classification/identification task on them.
While we attempt to identify gender from facial features, we are often curious about what features of the face are
most important in determining gender. Are localized features such as eyes, nose and ears more important or overall
features such as head shape, hair line and face contour more important?
There are a plethora of successful and robust face recognition algorithms on the web. Instead of using the inbuilt
tools that they provide, we start building various algorithms from scratch to gain a rich learning experience.
In this project, we will use the following methods of classification:
Eigenface Method
K-Means
GDA
SVM
Fisherfaces Method

We will be discussing these methods and their advantages along with limitations.

The Mathematical equations governing these methods are not discussed in this report.
 II. Application

There are several potential application areas where gender recognition would be used.
Human-Computer Interaction systems: These systems will giving more human like response if they will
be able to identify a humans attribute such as Gender.
Surveillance Systems: In smart surveillance systems it can assist in restricting areas to gender only, such
as train coach and hostel.
Biometrics: The face recognition system can be faster if it will be cut down to only a specific gender.
Targeted Advertising.

III. Challenges:
The image of a persons face exhibits many variations which may affect the ability of a computer vision system
to recognize the gender. We can categorize these variations as being caused by the human or the image capture
process.
Human factors are due to the characteristics of a person, such as age, ethnicity and facial expressions (neutral,
smiling, closed eyes etc.), and the accessories being worn (such as eye glasses and hat). Factors due to the
image capture process are the persons head pose, lighting or illumination, and image quality (blurring, noise,
low resolution). Head pose refers to the orientation of the head relative to the view of the image capturing
device. The human head is limited to three degrees of freedom, as described by the pitch, roll and jaw angles.
The impact of age and ethnicity on the accuracy of gender classification has been observed.
Benabdelkader and Griffin , after testing their classifier with a set of 12,964 face images, found that a
disproportionately large number of elderly females and young males were misclassified. In empirical studies by
Guo et al. using several classification method on a large face database, it was found that gender classification
accuracy was significantly affected by age, with adult faces having higher accuracies than young or senior faces.
In when a generic gender classifier trained for all ethnicities was tested on a specific ethnicity, the result was not
as good as a classifier trained specifically for that ethnicity.

IV. Data Set and Preprocessing:

The data set we will use is a set of high resolution color images of 396 female faces and 389 male faces
obtained from MUCT Database. All images are frontal view of face. The database provides diversity of lighting,
age and ethnicity.
The images also have variations in:
Head rotation and tilt.
Facial expression
Face/hair accessories
Position of face in image.
Preprocessing that may be applied to the face image include:
o Normalize for contrast and brightness (e.g. using histogram equalization)
o Removal of external features such as hair and neck region
o Geometric alignment (either manually or using automatic methods)
o Downsizing to reduce the number of pixels.
 V. Facial Feature Extraction
Feature extraction methods for face gender classification are categorized into geometric-based and appearance-
ased methods.
The Geometric-based method is based on measurements of facial landmarks. Geometric relationship between
these points are maintained but other useful information may be thrown away.
Appearance-based methods are based on some operation or transformation performed on the pixels of an image.
This can be done at the global (holistic) or local level. At the local level, the face may be divided into defined
regions such as eyes, nose and mouth or regularly spaced windows. In appearance-based methods, the
geometric relationships are naturally maintained, which is advantageous when the gender discriminative
features are not exactly known. However, they are sensitive to variations in appearance (due to view,
illumination, expression, etc.) and the large number of features.

VI. Related Work

Almost all of the works in gender classification involves extracting features from faces and classifying those
features using labeled data. They mostly differ in the way these two steps are performed. Therefore gender
classification approaches can be categorized based on the feature extraction and classification methods.
Early works in gender classification mostly used neural networks with face image as raw input. Some of these
are Golombs two-layer network called SEXNET, Tamura et al.s multilayer neural network, etc. Gutta takes a
hybrid approach using neural network and decision trees.
Moghaddam et al. in [6] uses non-linear SVMs to classify faces from low-resolution thumbnail images of size
21-by-12.

VII. Approach
In general, gender classification in supervised learning setting requires extraction of features from face-images,
training classifiers using those features and finally performing classification of new faces.
We will be using appearance-based approach with dimensionality reduction techniques for feature extraction.
The features extracted from training set are used for training and SVM classifier. And finally images in test set
are classified using classifier.
VIII.
IX.