A Non-Contact PPG Biometric System Based on Deep Neural Network Omkar R. Patil 1 , Wei Wang 2 , Yang Gao 2 , Wenyao Xu 2 , and Zhanpeng Jin 2 1 Binghamton University, SUNY, Binghamton, NY 13902 2 University at Buffalo, SUNY, Buffalo, NY 14260 opatil1@binghamton.edu, {wwang49,ygao36,wenyaoxu,zjin}@buffalo.edu Abstract The objective of this study is to develop a non-contact biometric system with photoplethysmogram (PPG). A novel method for non-contact PPG acquisition based on the Laplacian pyramid is proposed in this paper with the au- thentication module based on the deep neural network (DNN). Laplacian pyramid based video amplification tech- nique extracts the subtle changes of blood volume as a re- sult of the cardiovascular activities in the facial region. The video data was recorded from 20 subjects in varying light conditions at different places, resembling different sce- narios in the generalized environment. Authentication ac- curacy ranges from 66.67% to 100% with an average of 86.67%. In order to validate the repeatability of PPG wave- forms, a comparative analysis of the correlation coefficients for the waveforms recorded over a month are conducted. 1. Introduction In the today’s vastly growing world, biometric authen- tication and identification is one of the most important ar- eas in the security domain. The threats of malicious attacks on either the organizational databases or the personal se- curity systems can cause catastrophic financial loss as well as reputational damage. For instance, the access to smart home devices can create life threating events [31]. Sev- eral methodologies have been explored to protect personal information for centuries [24], such as word keys or mak- ing messages unintelligible to adversaries. The demands for more sophisticated security solutions is increasing with the advances in digital consumer electronic devices and its applications. The use of attributes that a person can carry (e.g., government ID or access cards issued by the respec- tive authority) or a person can remember (e.g., passwords, personal signatures, graphical patterns) has become popular in day-to-day life with increasing needs. These authentica- tion modules have serious disadvantages that can breach the 978-1-5386-7180-1/18/$31.00 c 2018 IEEE. 0 200 400 600 800 1000 Samples 6 8 10 12 Voltage(mV) 10 -3 ECG 0 20 40 60 80 100 120 Samples 0.5 0.6 Normalized Amplitude PPG Figure 1. Similarity between ECG and PPG waveforms for cardio- vascular activity system , e.g., i) information can be duplicated to provide access for others, ii) the attributes can be easily forgotten or stolen. The use of personal attributes such as face, retina, fingerprint, palm print that give unique personal patterns for person identity is known as biometrics [13]. The personal attributes that have been proposed to be used for biometric authentication and identification include: face [28], fingerprints [14],and retina [19]. The physiologi- cal or behavioral responses of a person in a day-to-day en- vironment or in a specific scenario can be a unique input for robust biometric systems, e.g., electrocardiogram (ECG) [11] and electroencephalogram (EEG) [18], etc. Recently, an emerging biometric method based on photoplethysmo- gram (PPG) [16, 31] has been proven to possess potential advantages over the conventional systems in terms of secu- rity, ease of use, universality, uniqueness, and cost. PPG is a technique being used to measure blood volume changes in proportion with the cardiac activities in the superficial atrial structure [2]. Figure 1 shows the morphological similarity between ECG and PPG in representing the cardiac activi- ties. The morphological characteristics of the PPG wave- form resemble the unique physiological properties of the