Gender Recognition from Images using Machine Learning

Abstract

Recognizing demographic traits of people, for example, age, gender and eth- nicity utilizing computer vision has been given great consideration in recent years. Such traits can be assumed to be an essential part in numerous appli- cations, for example, human-computer interaction, surveillance, content-based indexing and searching, biometrics, demographic studies and targeted adver- tising. The focus of this thesis is on gender recognition from human face im- ages using machine learning techniques. Experiments have been carried out on unconstrained and real life representative datasets like labeled faces in wild (LFW) and IMFDB. The First major contribution of the thesis is the use of a well known deep learning neural network ALEXNET for the task of gender recognition from human facial images. The other significant contribution is the proposal of IALEXNET. All the experiments are carried out with ALEXNET and proposed IALEXNET. Accuracy of the state-of-the-art technique for the task on hand on LFW is 94.81%. ALEXNET gives 96.12% and 93.52% accu- racy on LFW and IMFDB. Accuracy is further improved by IALEXNET and it is approximately 96.15% and 93.56% for LFW and IMFDB.