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

Abstract

Biometric authentication is a common way of granting access to a system or device. The ear, like fingerprints, retina, iris, face, voice, and so on, is a biometric modality. Compared to other biometric organs, the anatomy of a human's ear remains stable from birth to old life. As a visible organ with an easily acquired image, it may also be a source of a biometric signature that may be used to identify individuals. This research demonstrates two approaches to recognizing a person from 2D ear images: non-deep ML models and deep learning-based ML models. The first, or classic, model investigates computer vision preprocessing techniques such as converting an RGB image to monochrome, then rescaling and locating the entropy. The key weighted characteristics from the ear images were extracted using Independent Component Analysis (ICA) and Principal Component Analysis (PCA). A Gaussian Process Classifier (GPC) is then utilized for classification and several kernels such as RBF, Rational Quadratic, and Matern. In the second technique, a deep learning-based ML model called You Only Look Once (YOLO) is utilized to categorize the ear images and identify the source individual without preprocessing. We gathered a standard ear dataset (EarVN1.0 Dataset) from 164 people, totaling 27,592 training images. For testing reasons, 820 images were chosen randomly, five images from each of 164 people. The models were built on the Google Colaboratory server using the Python language framework and GPU-based implementation on the Jupyter Notebook.

Comparison of YOLO (V3,V5) and MobileNet-SSD (V1,V2) for Person Identification Using Ear-Biometrics

“…For ear detection and ear identification tasks, Faster-RCNN and VGG-19 are employed [37]. Recently, Shamim Akhter et al [19], [20] used the YOLOV3 model with a standard dataset named EarVN1.0. This collection was made up of 28412 ear pictures from 164 different groups.…”

“…The studies from [25], [26], [27], [28], [29] and [31] showed ML models for recognizing and identifying the ear from image data sets. However, [19], [20], [32], [37] studies use ear images to identify individuals. A typical classification issue requires the detection and localization of objects.…”

“…YOLOV5 [8] and MobileNetSSD [15] are used independently for detecting ear objects in person face images. In addition, YOLOV3 was evaluated in our previous works [19], [20] individually to recognize a person based on ear images. An effective and fast biometric security system must be able to identify and authenticate people based on their unique biological characteristics.…”

The Optic Disc Detection and Segmentation in Retinal Fundus Images Utilizing You Only Look Once (YOLO) Method

“…YOLO version one (YOLOv1) [37] as its neck in order to enhance information flow. With an enhanced bottom-up strategy, PA-Net leverages a unique feature pyramid network (FPN) design to improve the transmission of low-level features.…”

Modified Firefly Optimization with Deep Learning based Multimodal Biometric Verification Model

“…The commonly used biometric traits like voice, fingerprints, iris, face recognition, and palmprint necessitates the enrolment of such traits in databases for feature detection [3][4][5]. The biometric system acts as a auspicious and continuously developing technology utilized in the automatic systems for finding an individual effectively without remembering or carrying anything, such as passwords and Ids [6][7][8][9]. Numerous research works that the iris trait has higher advantages compared to other biometric systems relevant to the features such as fingerprint and face, this makes the iris to be generally adopted in several applications for high-reliability and precise biometric systems [9][10][11][12].…”