Multi-Feature Neural Network-Based Currency Authentication: Integrating Texture, Color, and Size for Robust Banknote Recognition

Abstract

This paper presents a multi-feature neural network-based system for banknote recognition, enhancing robustness and accuracy in challenging conditions such as worn, faded, and distorted banknotes. Texture features are extracted using Principal Component Analysis (PCA), while color information is combined with texture into a unified feature vector. This combined vector is then fed into a Multi-Layer Perceptron (MLP) neural network for classification. The system is evaluated on a dataset of 1,072 banknote images, including clean, faded, wrinkled, and dirty banknotes. The proposed method achieves 95% recognition accuracy, representing a 10% improvement over existing methods, particularly for distorted and worn banknotes. Experimental results demonstrate the effectiveness of combining texture, color, and size features for robust banknote recognition. This approach significantly improves the system's ability to handle discrepancies and challenges in real-world applications, such as ATMs and vending machines, ensuring reliable and real-time performance.