Automatic Race Estimation from Facial Images Using Shape and Color Features

Abstract

In this work, an attempt has been made to estimate the human race from facial images. This work is done for three major ethnicities: African, American, and Asian. The performances of the classifiers were tested with the faces images of African, American, and Asian population belonging to different age groups and genders. For this, first the facial part such namely the nose region is detected using the well-known Viola–Jones object detection technique. From the detected nose, 9 Zernike moments, 81 HoG features, and 2 color features are extracted to estimate the race using classifiers namely ANN and SVM. Four hundred and fifty sample images taken from the FERET database were considered for this study, out of which 330 images were used for training and for 120 testing. The accuracy of the model obtained through artificial neural network is 91.06%, whereas the accuracy obtained by applying SVM is 95.1%. From the results obtained, it is evident that SVM outperforms ANN in identifying the race of a person from his/her facial image and could be effectively employed in automatic race estimation systems based on facial images.