A Real-time Machine Learning-Based Person Recognition System With Ear Biometrics

Abstract

Biometric authentication is a very popular method for authorizing a person to a system or a device. The ear is a biometric modality such as fingerprints, retina, iris, face, voice, etc. A person's ear structure remains the same from early childhood to old age as compared to other biometric organs in the human body. Thus, the ear can also be a source of a biometric pattern to identify a person, since it is a visible organ and its image can be easily captured. In this paper, we present two different approaches, namely, Non-Deep ML models and Deep Learning-based ML model for identifying a person from 2D ear images. The first or traditional model explores computer vision preprocessing tactics including the conversion of the RGB image to grayscale, then rescaling and finding the histogram. Independent Component Analysis (ICA) and Principal Component Analysis (PCA) were used to extract the major weighted features from the ear images. Then, for classification, a Gaussian Process Classifier with different kernels including RBF, Rational Quadratic, and Matern is applied. In the second approach, a deep machine learning (ML) algorithm namely You Only Look Once (YOLO) is used to classify the ear images, without any preprocessing, and identify the source person. We collected a standard ear dataset (EarVN1.0 Dataset) from 164 individuals, with a total of 27,592 training images. Randomly 820 images, 5 images of each 164 persons are selected for testing purposes. The models were implemented using the python language framework and GPU-based implementation on the Jupyter Notebook at the Google Colaboratory server.