Improving Performance of PNN using Clustered ICs for Gender Classification

Abstract

The research presented in this paper proposes a novel gender classification approach using face image. The approach extracts features from grayscale face images through Infomax ICA and subsequently selects features using k-means clustering and classifies the clustered features employing PNN. All the experimental evaluations are done on cropped face images from FERET database using 280 faces for training and 120 different faces for testing. The approach, when features are not clustered gives maximum accuracy of 93.33%. However the proposed approach yields 95% accuracy through employing clustering on features, which is significant for gender classification using low resolution (118 × 97) face images.