Periocular Gender Classification using Global ICA Features for Poor Quality Images

In recent years, the research over emerging trends of biometric has grabbed a lot of attention. Periocular biometric is one such field. Researchers have made attempts to extract computationally intensive local features from high quality periocular images. In contrast, this paper proposes a novel approach of extracting global features from periocular region of poor quality grayscale images for gender classification. Global gender features are extracted using independent component analysis and are evaluated using conventional neural network techniques, and further their performance is compared. All relevant experiments are held on periocular region cropped from FERET face database. The results exhibit promising classification accuracy establishing the fact that the approach can work in fusion with existing facial gender classification systems to help in improving its accuracy.