Facial Image Classification Based on Age and Gender

Abstract

Automatic face image classi cation based on facial features is identi ed as a challengeable task in many computer vision systems which can be used in several application areas. In this thesis we will introduce a novel approach to solve the challengeable problem of facial image classi cation based on age and gender. The approach uses geometric variations of facial features according to the gender and age groups and uses a neural network based classi er for estimate the gender and age of a given facial image. We use four age groups including age ranges of 8 { 13, 14 { 25, 26 { 45 and 46 { 63 due to di culty of estimating an exact age for the given facial image. Whole methodo- logy of classi cation includes three basic steps including preprocessing, feature extraction and classi cation. Feature extraction is the most important step in the methodology which identi es the facial feature points from eyes, nose, mouth, forehead area, eye lid area and cheek area to calculate the parameters need for classi er. These feature points are located using projection functions and using Sobel gradient magnitude. Wrinkle features are ob- tained to classify a given face image into one of the gender groups and geometric features are calculated to classify the images according to the gender. First an image is classi ed into one of the gender classes and then nds out the corresponding age group for the given image. Proposed methodology is trained using data taken from 550 images and the results were checked using 40 test images. Results achieved for the gender classi cation is 85% for the test data set and 87.5% for the training data set. For the age classi cation it is 75% for the training data set and 60.86% for the test data set. The proposed methodology gives a 56.52% overall classi cation for the given test data set while classi cation rate is 66.45% by the human for the similar experiment.