Classification of Face Images Based on Gender using Dimensionality Reduction Techniques and SVM Fahim Mannan 260 266 294 School of Computer Science McGill University Abstract This report presents gender classification based on facial images using dimensionality reduction techniques such as Principal Component Analysis (PCA) and Independent Component Analysis (ICA) along with Support Vector Machine (SVM). The input dataset is divided into training and testing dataset and experiments are performed by varying dataset size. The effect of performing image intensity normalization, histogram equalization, and input scaling are observed. The outcomes of the experiments are analogous to published works that apply similar techniques. 1. Introduction Gender classification using facial images has been of interest for quite some time. Early works were mostly related to psychological research where the process by which humans determine gender from faces is studied. Humans are very good at determining gender from facial images. Even if the face is cropped to remove all gender cues, we can identify gender with very high accuracy ([1]).More recently automated gender classification from facial images has gained much interest in the computer vision and machine learning community. This is because of its extreme importance in Human Computer Interaction, demographic research, and security and surveillance applications. It can also augment other important areas like face recognition, age and ethnicity determination. Several approaches have been taken to classify facial images based on gender. This report addresses one particular approach using dimensionality reduction (ICA and PCA) and Support Vector Machine (SVM). One of the challenges of automatic gender classification is to account for the effects of pose, illumination and background clutter. Practical systems have to be robust enough to take these issues into consideration. Most of the work in gender classification assumes that the frontal views of faces, which are pre-aligned and free of distracting background clutters, are available. Towes et al.[2] provides a framework that is free of these assumptions and can classify faces by first automatically detecting, then localizing and finally extracting features from arbitrary viewpoints. But in this project, only the facial images with full frontal views are considered. The report is organized as follows - section 2 presents an overview of related research. Section 3, looks at the general approach and explains dimensionality reduction techniques as well as SVMs. Experiments are illustrated in section 4 and the results are discussed in section 5. Section 6 concludes with a discussion of possible future works.