Computing Conference 2017 18-20 July 2017 | London, UK A Graph Matching Algorithm for User Authentication in Data Networks using Image-based Physical Unclonable Functions Ali Valehi 1 , Abolfazl Razi 1 , Bertrand Cambou 1 , Weijie Yu 2 , and Michael Kozicki 2 1 School of Informatics, Computing and Cyber Security (SICCS) Northern Arizona University, Flagstaff, AZ 86011 2 School of Electrical,Computing and Energy Engineering (ECEE) Arizona State University, Tempe, AZ 85281 Abstract—Recently, Physically Unclonable Functions (PUFs) received considerable attention in order to developing security mechanisms for applications such as Internet of Things (IoT) by exploiting the natural randomness in device-specific characteris- tics. This approach complements and improves the conventional security algorithms that are vulnerable to security attacks due to recent advances in computational technology and fully automated hacking systems. In this project, we propose a new authentication mecha- nism based on a specific implementation of PUF using metal- lic dendrites. Dendrites are nanomaterial devices that contain unique, complex and unclonable patterns (similar to human DNAs). We propose a method to process dendrite images. The proposed framework comprises several steps including denois- ing, skeletonizing, pruning and feature points extraction. The feature points are represented in terms of a tree-based weighted algorithm that converts the authentication problem to a graph matching problem. The test object is compared against a database of valid patterns using a novel algorithm to perform user identi- fication and authentication. The proposed method demonstrates a high level of accuracy and a low computational complexity that grows linearly with the number of extracted points and database size. It also significantly reduces the required in-network storage capacity and communication rates to maintain database of users in large-scale networks. Keywords—Information-Security; Image-Processing; Graph- Matching; Authentication I. I NTRODUCTION Image detection is commonly used as a non-secret iden- tification method because cloning an image is hard to pre- vent, thereby the use of an additional secret password or a private key is important to ensure trustworthy authentication. The secret keys or passwords are commonly stored in non- volatile memory of secure microcontrollers such as electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM) and Flash memory. This approach is costly both in terms of key distribution and design procedure [1]. Moreover, security keys that are stored as digital information in memory are vulnerable to invasive attacks by adversaries who are constantly looking for ways to penetrate the security system and retrieve keys. Utilizing Physically Unclonable Functions (PUF) is a new emerging security technique that has gained attention recently as a central building block in variety of cryptographic proto- cols. PUF technology leverages the natural randomness gener- ated during the manufacturing of microelectronic components such as SRAM memories, ring oscillators, and gate delays. In this approach, a secret key which is required to produce a Challenge-Responses-Pair (CRP) for an authentication session is generated by PUF as a function of its hidden, hard-to-access and unique characteristics. A reference key is stored for each user in a secure databased in the network. An authentication process is deemed positive when the generated response by PUF matches the one in the network. In addition to the requirement of high randomness of keys generated by PUF to maximize the entropy of the cryptographic authentication process, PUFs should be resistant to side channel attacks. In this paper, we are investigating new methods to use the image of unclonning materials to produce secret PUFs that add an additional security level to the system and improve the current identification and trustworthy authentication methods. Fig. 1. A sample dendrite object. The dendrite wafer panel includes 24×24 = 576 dendrite objects. 978-1-5090-5443-5/17/$31.00 c 2017 IEEE 863 | Page