Abstract
The degree of utilization of Quick Response (QR) codes is sharply increasing due to the wide availability of smart devices. The primary purpose of the QR code is to ensure that an extensive message is fully transferred in a compact data format. Like any environment, security is an essential issue where QR codes are utilized. Such problems include the lack of signing information in a QR. This study aims to exploit the QR code hiding mechanism without spoiling the value of the code in the QR code while determining it using several machine learning algorithms. Consequently, several new QR image datasets are generated with varying sizes and variations to examine the classification of the proposed message-hiding scheme. This study used state-of-the-art models (VGG16, Xception) and a CNN-based model for QR code classification but only achieved 50% accuracy across four QR code dataset variants. Unsatisfied with these results, the study then employed the histogram feature density technique with various machine-learning (Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF)) and deep learning (DL) models. The experimental results reveal that adapting the histogram density method in the proposed scheme for feature creation achieved an overall success rate of approximately 99.98%. Moreover, the study further aims to simulate single-layer QR codes from hackers’ perspective that pretends to look like two-layer QR code systems. As a result of this simulation study, the performance was tested using different classification algorithms. In most cases, except for one, the DL model performed better by attaining a success rate above 90%.
Get full access to this article
View all access options for this article.