JECET; June – August-2013; Vol.2.No.3, 876-882.
Journal of Environmental Science, Computer Science and Engineering & Technology
An International Peer Review E-3 Journal of Sciences and Technology
Available online at www.jecet.org Computer Science
Research Article
JECET; June – August 2013; Vol.2.No.3, 876-882. 876
An Elegant Neural Network based draw near for currency Recognition
Rumi Ghosh1 and Rakesh Khare2
1Department of Computer Science and eng., Raipur Institute of Technology, Raipur, India
2Department of Information Technology, Raipur Institute of Technology, Raipur, India
Received: 29 July 2013; Revised: 16 August 2013; Accepted: 21 August 2013
Abstract: Currency recognition is an image processing technology that is used to identify currency of various countries. The requirements for an automatic banknote recognition system have offered many researchers to build up robust and dependable techniques. A reliable currency recognition system is important for the automation in different sectors for a country. The focus of most of the conventional currency recognition systems and machines is on recognizing counterfeit currencies but it is not enough for practical businesses. Reliable Paper currency recognition systems should be able to recognize banknotes from each side and each direction. In this paper we have represented a currency recognition system using neural network that uses histogram based feature extraction and multilayer Perceptron model for classification.
Keywords: Image Processing, Neural Network, Feature Extraction, Multilayer Perceptron
JECET; June – August 2013; Vol.2.No.3, 876-882. 877 INTRODUCTION
Modernization of the financial system is a milestone in protecting the economic prosperity, and maintaining social harmony. There are approximately more than 150 currencies all over the world, each of them looking totally different. Automatic machines capable of recognizing banknotes are massively used in automatic dispensers of a number of different products, ranging from cigarettes to bus tickets, as well as in many automatic banking operations. By expansion of modern banking services, automatic schemes for paper currency recognition are significant in many applications. The requirements for an automatic banknote recognition system have offered many researchers to build up robust and dependable techniques. Speed and precision of processing are two vital factors in such systems. The technology of currency recognition aims to search and extract the visible and hidden marks on paper currency for efficient classification.
Currency recognition is an image processing technology that is used to identify currency of various countries. Probabilities that the paper currencies of various countries are probably interweaved together therefore rises increasingly. It is a challenge for conventional paper currency recognition systems. However, the focus of most of the conventional currency recognition systems and machines is on recognizing counterfeit currencies. It is not enough for practical businesses. The reason is that in most of banks, especially those internationalized banks, there are large quantities of cash belonging to many different countries need to be process, and it is possible that all of them are real cashes. The situation that cashes belonging to different countries mixes together is possible to occur. It cannot be processed with conventional currency recognition systems.
Paper currency recognition systems should be able to recognize banknotes from each side and each direction. Since banknotes may be defaced during circulation, the designed system should have a meaningful accuracy in detecting torn or worn banknotes. The technology of currency recognition is used to research the visible and hidden currency characters, identify features all-around and dispose of the process on time. The original information has a loss because paper currency will get to the worm and blurry, even damaged by human being in circulation.
Related Work: In1 proposed a Euro banknote recognition system using two types of neural networks;
a three-layered perceptron and a Radial Basis Function (RBF) network. In this paper, author has proposed a banknote recognition system composed of two parts; a classification part and a validation part. The classification part uses a three-layered perceptron and the validation part uses several RBF networks. While the three-layered perceptron is a well known method for pattern recognition and is also very effective for classifying banknotes, it makes boundaries only for classifying given training data and is unassured of rejecting unknown data. The RBF network has a data approximation property, which seems a proper tool for rejecting unknown data. It is able to configurate the system employing only one RBF network.
In2 the proposed system “SLCRec” comes up with a solution focusing on minimizing false rejection of notes. Sri Lankan currency notes undergo severe changes in image quality in usage. Hence a special linear transformation function is adapted to wipe out noise patterns from backgrounds without affecting the notes’ characteristic images and re-appear images of interest. The transformation maps the original gray scale range into a smaller range of 0 to 125. Applying Edge detection after the transformation provided better robustness for noise and fair representation of edges for new and old damaged notes. A three layer back propagation neural network is presented with the number of edges detected in row order of the notes and classification is accepted in four classes of interest which are 100, 500, 1000 and 2000 rupee notes
JECET; June – August 2013; Vol.2.No.3, 876-882. 878 In3 presented a currency recognition system using ensemble neural network (ENN). The individual neural networks (NNs) in an ENN are trained via negative correlation learning. The objective of using negative correlation learning (NCL) is to expertise the individuals on different parts or portion of input patterns in an ensemble. The image of different types note is converted in gray scale and compressed in the desired range. Each pixel of the compressed image is given as an input to the network. This system is able to recognize highly noisy or old image of TAKA. Ensemble network is very useful for the classification of different types of currencies. It reduces the chances of misclassification than a single network and ensemble network with independent training.
In4 put forward a new image based technique for Birhani paper currency recognition based on two classifiers, the weighted Euclidean distance using suitable weights and the Neural Network. First of all color image of paper currency having quality approximately equal to 600 dpi is obtained through scanning process. In pre-processing step four different kinds of images are obtained from color image, viz. the binary image; the gray scale image using Sobel mask; the gray scale image using Prewitt mask; and the gray scale image using Canny mask. Then features are extracted by calculating the sum of pixels of each of the four images. Also, the Euler number is calculated for each of the images, then computed the correlation coefficient of input image after converting it to gray scale. After feature extraction paper currency classification is done by using two different methods called Weighted Euclidean Distance (WED) and Neural Networks using feed forward back propagation.
METHODOLOGY
Our Proposed Currency recognition system divided into following parts:
1. Pre-Processing of currency image.
2. Feature Extraction.
3. Classification
Fig: 1: Proposed Methodology
Image pre-processing: Pre-processing of image are those operations that are normally required prior to the main data analysis and extraction of information. Here image resizing is performed because the currency image is too large to process.
JECET; June – August 2013; Vol.2.No.3, 876-882. 879 Feature Extraction: Feature extraction of images is challenging work in digital image processing.
The feature extraction of Indian currency notes involves the extraction of features of Indian currency.
During the feature extraction process the dimensionality of data is reduced. This is almost always necessary, due to the technical limits in memory and computation time. Extracting too many features will not only increase the cost but also sometimes lower the system performance in terms of execution time. Therefore, we have to choose only the critical features that are easy to extract but difficult to imitate. Basically, at first instance, people may not pay attention to the details and exact characteristics of banknotes for their recognition, rather they consider the common characteristics of banknotes such as the size, the background color (the basic color), and texture present on the banknotes. In this method, these characteristics will be used to differentiate between different banknote denominations.
• Size: The first step of recognition of algorithm proposed by Hassanpour5 , considers size of the banknote. It is considered because during circulation of banknote worn and torn reduces its size, also it may increased slightly by rejoining torn banknote. Therefore they proposed decision tree as follows
|ݔ−ݔ0| <݀ݔ & |ݕ−ݕ0| <݀ݕ .. (1) Where x0 and y0 are size of the testing paper currency, x and y are size of reference paper currency. dx, dy shows changes in the vertical and horizontal directions.
• Texture Feature: For texture feature LBP (Local Binary Pattern) operator6 is used. Texture is the visible feature of the paper currency.
LBP operator: It is originally introduced by Ojala et al6 . In LBP, the neighborhood pixels are converted to binary code 0 or 1 by using the gray value of the centre pixel as threshold and further arranged to form as a ordered pattern. The feature extracted with LBP gives the relationship of the texture within local area. LBP code for pixel p is defined as:
ܮ( ) = ...(2) Where
gp – Gray value of the centre pixel p.
gi – Gray value of the ith pixel, 8-neighbourhood of p
i – 0, 1 ...7. s(t) = gi - gp, is the threshold function and given by : ݏ (ݐ)= 1, ݐ≥0
0, ݈݁ݏ݁ ...(3) From equation 1 we can say that LBP can produce 256 kinds of different outputs, corresponding to 256 kinds of different binary patterns.
• Colour Feature: If the primary image is in RGB format, then after resizing it is converted to HSV colour space7. Advantage of HSV colour space is that it is closer to human conceptual understanding of colours and has ability to separate chromatic and achromatic Components.
Feature extraction of a colour image can be done by analysing its colour histogram, hue, saturation, intensity (or value) In HSV (Hue, Saturation, Value) space hue distinguishes colour, Saturation is the percentage of white light added to a pure colour and value represents perceived light intensity
JECET; June – August 2013; Vol.2.No.3, 876-882. 880 Classification: After getting features of currencies, it is essential to recognize the pattern of the currencies on the base of these features, which should be practiced by an effective recognition system called classifier. A Neural network based recognition scheme is used here for currency recognition.
A multi-layered perceptron is a kind of feed forward neural networks which is well known tool for pattern recognition. In our system, a three-layered perceptron is employed in the classification part.
The structure of a three-layered perceptron is shown in Fig. 2.
Colour
Size
Texture
1 Rs.
2 Rs.
5 Rs.
10 Rs
20 Rs.
50 Rs.
100 Rs.
500 Rs.
Xi
Wjk Zj
Vij
Yj
Fig 2: Structure of Multilayer Perceptron
A three-layered perceptron is composed of an input layer, a hidden layer and an output layer. Each input neuron is fully connected to the hidden neurons and each hidden neuron is fully connected to the output neurons. The strength of the connection between neuron i and j is represented by weight value wij . The output of each neuron is calculated by the sigmoidal function and the input of the sigmoidal function is the sum of products of the output values and the weight values from the previous layer.
The output of the neuron j and a sigmoidal function are represented by
where f(x) is the sigmoidal function and yj is the output of the neuron j.
The input value of each input neuron corresponds to a component of an input vector x and each output neuron corresponds to each class index.
JECET; June – August 2013; Vol.2.No.3, 876-882. 881 RESULTS
Fig. 3 gives the resultant values of MSE (Mean Square Error) and %E obtained after training, testing and validation. Training of the network takes 148 epochs in total. Fig. 4 shows the performance of network for each training, testing and validation. The best validation performance is achieved at epoch 36. It is clear from the figure that 97.34% correct recognition has been achieved which is quite encouraging. So, there is only 2.66% misclassification.
Fig. 3: Results after Training, Testing and Validation of NN
Fig. 4: Performance of Neural Network
Fig. 5: Snapshot of Currency Recognition System
JECET; June – August 2013; Vol.2.No.3, 876-882. 882 S.No. Currency Denomination Images Correctly
reocognized/total no. of images
Recognition rate (in %)
1 1 Rs. 75/78 96.1
2 2Rs. 76/78 97.4
3 5Rs. 33/34 97
4 10 Rs. 40/46 86.9
5 20 Rs. 43/46 93.4
6 50 Rs. 66/69 95.6
7 100 Rs. 79/83 95.1
8 500 Rs. 28/29 96.5
CONCLUSION
With development of modern banking services, automatic methods for paper currency recognition become important in many applications such as in automated teller machines and automatic goods seller machines. The technology of currency recognition aims to search and extract the visible and hidden marks on paper currency for efficient classification. We have proposed currency recognition system for verifying Indian paper currency. The system has a good performance for both accepting valid banknotes and rejecting invalid data.
REFERENCES
1. Masato Aoba, Tetsuo Kikuchi, Yoshiyasu Takefuji, “Euro banknote recognition system using a three layer perceptron and RBF networks”, IPSJ Transaction on Mathematical Modeling and Its Application, Vol 44, No. SIG 7 (TOM 8), May 2003, Pp. 99-109.
2. D. A. K. S. Gunaratna, N. D. Kodikara and H. L. Premaratne, “ANN Based Currency Recognition System using Compressed Gray Scale and Application for Sri Lankan Currency Notes-SLCRec”, in proceedings of world academy of science, engineering and technology, Nov 2008,vol. 35,ISSN 2070- 3740, Pp. 235-240.
3. Kalyan Kumar Debnath, Sultan Uddin Ahmed, Md. Shahjahan, “A Paper Currency Recognition System Using Negatively Correlated Neural Network Ensemble”, Journal Of Multimedia, December 2010, Vol. 5, No. 6, Pp. 560-567.
4. Ebtesam Althafiri, Muhammad Sarfraz, Muhannad Alfarras, “Bahraini Paper Currency Recognition”
Journal of Advanced Computer Science and Technology Research, June 2012, Vol. 2 No.2, Pp. 104- 115
5. H. Hassanpour, A. Yaseri, G. Ardeshiri, Feature Extraction for Paper Currency Recognition, IEEE Transactions, 1-4244-0779-6/07, 2007.
6. T Ojala,M Pietikainen, and D Harwood, A comparitive study of texture measures with classification based on feature distributions, Pattern recognition, Vol.29, No. 1, 1996, pp.51-59.
7. G. Trupti Pathrabe, Mrs. Swapnili Karmore, A Novel Approach of Embedded System for Indian Paper Currency Recognition, International Journal of Computer Trends and Technology- May to June Isue 2011, ISSN: 2231-2803.
*Corresponding Author: Ms.Rumi; 2H.O.D Department of Information Technology, Raipur Institute of Technology, Raipur , India
