20 September 2013 On the design of embedded solutions to banknote recognition
Author Affiliations +
Optical Engineering, 52(9), 093106 (2013). doi:10.1117/1.OE.52.9.093106
Abstract
Banknote recognition systems have many applications in the modern world of automatic monetary transaction machines. They are traditionally based on simple classifiers applied over manually selected areas. A new solution in this field, borrowed by content-based image retrieval (CBIR), which is based on dense scale-invariant feature transform features in a bag-of-words framework followed by a support vector machine (SVM) classifier, is explored. The proposed computer vision system for banknote recognition, on one hand, enables recognition at high accuracy and speed, and, on the other hand, provides basis for further applications, e.g., counterfeit detection and fitness test. This approach makes the system robust to various defects, which may occur during image acquisition or during circulation life of banknote. We implemented and tested on an embedded platform three state-of-the-art classification techniques [SVM, artificial neural network (ANN), and hidden Markov model (HMM)]. The comparative results are reported for accuracy with different sizes of the training datasets and with various types of datasets. In this framework, the SVM classifier outperforms ANN and HMM on the basis of speed and accuracy on our embedded platform.
© 2013 Society of Photo-Optical Instrumentation Engineers (SPIE)
Adnan Rashid, Andrea Prati, Rita Cucchiara, "On the design of embedded solutions to banknote recognition," Optical Engineering 52(9), 093106 (20 September 2013). http://dx.doi.org/10.1117/1.OE.52.9.093106
JOURNAL ARTICLE
12 PAGES


SHARE
KEYWORDS
Digital signal processing

Sensors

Feature extraction

Image processing

Neurons

Optical engineering

Image resolution

Back to Top