Human Recognition using Ear based Deep Learning Features

Recognition of person on the basis of his/her ear features is an upcoming area of research in the biometric field. In compression to other biometric traits, datasets with large number of images is not available and as such performance gains due deep learning methodology is still unexplored. In this paper aggressive data augmentation is carried out to artificially increase the number of samples in the AMI ear database which has only seven images per class. The augmented dataset is subjected to feature extraction by deep learning and classification by a classical model to obtain a hybrid deep learning classical model. The deep learning architecture inceptionV3 is used as feature extraction mechanism in which input to avgpool layer is taken as feature vectors and the dimensions of the feature vectors are reduced using principal component analysis. The dimensionality reduced features are classified using linear and quadratic classifier and best accuracy rate of 98.1% is obtained on quadratic SVM classifier.