학술논문

Texture Analysis by Deep Twin Networks for Paper Fraud Detection
Document Type
Conference
Source
2022 30th Signal Processing and Communications Applications Conference (SIU) Signal Processing and Communications Applications Conference (SIU), 2022 30th. :1-4 May, 2022
Subject
Communication, Networking and Broadcast Technologies
Computing and Processing
Fields, Waves and Electromagnetics
Power, Energy and Industry Applications
Robotics and Control Systems
Signal Processing and Analysis
Training
Iris
Measurement uncertainty
Signal processing
Fingerprint recognition
Feature extraction
Fraud
texture
Siamese Networks
paper
fingerprinting
hypothesis testing problem
Language
Abstract
This study proposes a method to distinguish fake documents from the originals using the textural structures of the papers they are printed on. The study is based on observations showing that paper textures are different and unique, just like fingerprint and iris tissue. This method, which captures the visually distinctive features of paper textures, can detect whether the documents of which the origin is suspected are fake or not. The proposed method can measure Type-2 error by training a Siamese network and thresholding the similarity results between two papers. Experimental results show that the proposed method has better distinguishing features than classical methods.