Source Verification of Printed Logos for Anti-Counterfeiting Applications

With the advent of the digital era, the growing prevalence of counterfeit goods poses serious threats to consumer safety and brand integrity, necessitating the development of novel techniques to combat forgery and counterfeiting. This paper focuses on the authentication of printed logos. The problem is formulated within a verification framework and is addressed using three distinct Siamese Neural Network (SNN) architectures: two shallow-and-wide networks and one based on the Xception model. Specifically, we investigate three SNN variants: Multiple Convolutions Summation (MCS-SNN), Multiple Convolutions Concatenation (MCC-SNN), and Mini-Xception SNN. To enhance authentication effectiveness, we adopt a two-step authentication process, beginning with an initial coarse analysis based on chromatic features, followed by a detailed verification that leverages micro-geometric, printer-specific artifacts, with each step focusing on distinct authentication patterns. The results obtained in both closed and open-set conditions indicate promising authentication accuracies, demonstrating the effectiveness of our approach for robust identity verification in real-world scenarios. Furthermore, the lightweight design of these models underscores their practical suitability for deployment on consumer-grade devices, highlighting their potential for real-world anticounterfeiting applications.