Paper Title (use style: paper title)

0 downloads 0 Views 2MB Size Report
The author(s) can distribute this PDF file for research and educational (nonprofit) purposes only. Distribution ... denomination, image processing, otsu thresholding. I. INTRODUCTION. Paper currency recognition (PCR) is paramount concept in.
A Robust Method for Identification of Paper Currency Using Otsu’s Thresholding Muhammad Ahsan Ansari, Suresh Kumar Mahraj Department of Computer Systems Engineering Mehran University of Engineering and Technology Jamshoro, Pakistan. [email protected], [email protected]

Abstract— In this modern world, the correct identification of paper currency is an essential need for the automation of current systems. The automatic currency identification method is highly required to reduce the human power in recognizing the currency and its associated value. Sometimes, the damaged or blurred currency and the complex design of the currency make the task of identification much difficult. In this paper, a method is proposed that can identify paper currency by considering the identification marks given on the currency. Specifically, this study focuses on the identification of Pakistani paper currency. The proposed method makes use of the fundamental image processing techniques for differentiating the major denominations. Particularly, the proposed algorithm uses Otsu’s thresholding method and circulatory matrix criteria for locating the special identity marks given on the currency. The effectiveness of the proposed approach has been shown by performing various experiments on all of the major denominations of Pakistani paper currency with varying intensities. The Proposed method shows quite satisfactory results in terms of accuracy and efficiency. Keywords— currency recognition, region denomination, image processing, otsu thresholding.

of

interest,

I. INTRODUCTION Paper currency recognition (PCR) is paramount concept in correctly identifying the currency and its corresponding value [1]. Manual identification of currency requires expertise of the examiners which involves human labor and time. To facilitate the management of economy of countries, the idea of automatic currency recognition is of great interest for researchers [2]. Usually, most of the countries have banknotes in the form of either bills or coins and the recognition of both of these is critical for various applications that may include the assistance of blind persons, machines for counting bank notes, classification of money deposited in automatic teller machines [3]. Bill and coins possess different properties concerning recognition. Electronic banking, money exchange, currency monitoring system is some of the potential applications of PCR. Roughly, more than 200 currencies are being used by different countries in all over the world [4]. Almost all of the currencies used all over the world look different and hence have different features. For example, size of the paper, identification marks, color, pattern and many other. The task for distinguishing the different currencies is not an easy job especially for the person who works at the places like money

exchange offices. Remembrance of the symbols for each of the currencies is quite difficult task and also may lead to wrong identification. Moreover, due to the advancements in imaging technologies for scanning and printing, the production of counterfeit banknotes has become much easier and hard to detect [5]. Therefore, an automatic system is highly required that can identify the given currency accurately with high efficiency. All of the visible features present on paper currency play a vital role in their recognition. From previous studies it has been found that the best way for identifying a note is to locate the visible features of the note [6]. II. LITERATURE REVIEW Like any other object recognition such as vehicle’s license plate, iris of the eyes, text in document, sign in sign language and many others, the task of identifying the currency is essential. Various researchers have given their contributions in the field of currency identification and recognition [7-13]. Primarily, most of the existing literature techniques make use of image processing and neural network based methods [9-13]. The method in [14] has performed the currency identification task but by using a large number of features as compared to the approach presented in this paper. The currency recognition method described by the authors in [9] is based on the neural network where symmetrical masks are used to consider particular signs in paper currency. Another method has been described in [15] where quite different feature set is used but their method is specific to the regional currency color and marks. In another study [16], three characteristics (size, color, and texture) of the paper currency are utilized for the currency recognition problem. In [17] infrared light is used to expose the intaglio printing of currency note, which is the embossed or raised printing on the currency for identifying it by touch. In their approach, the actual infrared images are embossed and converted to binary form prior to apply template matching because the contrast of the infrared images is not too good. Their method attains a good accuracy measure but at the expense of requiring embossing and other such complex techniques. On the contrary, the proposed method is simple yet efficient in identifying the Pakistani paper currency. The simplicity of the proposed method lies in the fact that it makes use of the basic image processing techniques in an efficient manner.

The usage of this PDF must comply with IEEE ICSCEE Provisions on copyright. The author(s) can distribute this PDF file for research and educational (nonprofit) purposes only. Distribution by anyone other than author(s) is prohibited.

More specifically, the proposed method uses circulatory matrix criteria for locating the special marks given on the currency for identification. The major denominations of Pakistani papers notes have been shown in Fig. 1. By looking at the different denominations of Pakistani notes given in Fig. 1, it is observed that the color of the currencies changes almost uniformly within a note. This uniformity of intensities compelled us to use a global thresholding method [18] in order to find the binarized version of the input image (paper currency). Moreover, the regions enclosed by the red bounding boxes in Fig. 1 represents the region of interest (ROI) for the proposed algorithm which specifies the special marks that can be used for identification of particular currency. As can be observed, these special marks are unique in all of the denominations of Pakistani currency. III. PROPOSED METHOD Pakistani paper currency consists of seven major denominations (PKR-10, PKR-20, PKR-50, PKR-100, PKR500, PKR-1000, and PKR-5000) as described in Fig. 1. According to the data mentioned on the official website of State Bank of Pakistan, the denomination PKR-10 comprises of green color, PKR-20 comprises of orange-green color, PKR-50 consists of purple color, PKR-100 consists of red color, PKR500 consists of rich deep green color, PKR-1000 consists of dark blue color, and PKR-5000 comprises of mustard color. The block diagram describing the overall process of the proposed method is presented in Fig. 2.

Input

Image Acquisition

Output

Binarization Using Otsu’s Thresholding

Extracting ROIs

Display the Identified Currency D i ti

Computation of Circulatory Matrix for each ROI

Fig. 2. Block Diagram of the proposed method

The proposed algorithm identifies the robust feature of the Pakistani currency notes for the identification of particular note. All of these notes in Fig. 1 can be differentiated uniquely on the basis of special identity marks given on the bottom-left of each note. After analysis of these paper currency notes it is found that all these special identity marks fall into three categories: 1) rectangular object, 2) circular object, and 3) no identity marks. The note of denominations 20, 50, and 100 consists of one, two and three rectangular objects; (special identity marks) respectively and hence may fall in category-1. The note of denominations 500, 1000, and 5000 consists of one, two and three circular objects, (special identity marks) respectively and hence may fall in category-2. Whereas, the note of PKR-10 consists of no special identity marks and hence fall in third category. This special case can be identified easily by the absence of the identity mark. As per the data of State bank of Pakistan, the dimensions of each of the seven denominations are defined in Table 1, where height and width of all the notes are mentioned in millimeter. It can be noticed from the values given in Table 1 that the height of each note is same that is 65mm whereas the width varies according to the denomination. By considering all this information regarding currency notes; an algorithm is proposed that efficiently identifies the given currency note and displays its value as well in an automatic fashion. The algorithm of the proposed method is described below. TABLE I.

Fig. 1. Seven Denominations of Pakistani Paper Currency

BANKNOTES DIMENSIONS

Currency Value

Height (mm)

Width (mm)

Color

10

65

115

Green

20

65

123

Brown / Orange Green

50

65

131

Purple

100

65

139

Red

500

65

147

Rich Deep Green

1000

65

155

Dark Blue

5000

65

163

Mustard

Algorithm: Step-1: Read/Capture the currency image of any denomination. Step-2: Binarize the input image using Otsu’s thresholding. Step-3: Extract the Region of Interest (ROI). Step-4: Compute the circulatory matrix of the ROI. Step-5: Display the identified currency denomination. The following sub-sections explain the steps of the proposed algorithm in detail. A. Image Aquisition The first step of the proposed algorithm is inputting an image to the system which can be acquired by either a scanner or a digital camera. The picture should be captured or scanned in a way as shown in Fig. 1. A very light intensity image may lead to inaccurate results. B. Pre-processing The pre-processing step is applied in order to change the nature of the inputted image. In the proposed methodology, the input image is converted into its equivalent binarized image so that the details and the features of the image are more enhanced and visible. During pre-processing stage, noise may also be removed using basic image processing methods. For example, simple mean and median filtering may be used to filter noise. For binarization, the most famous method of global thresholding that is Otsu’s thresholding has been used specifically. Otsu thresholding algorithm automatically selects optimal global threshold for an image by working on the assumption that pixels of an image has two classes or has bimodal histogram. In Otsu thresholding, a threshold is exhaustively searched that minimizes the variance within the classes. The intra-class variance is defined by (1) as weighted equation of variances of each class: = σ w2 (t ) q1 (t )σ 12 (t ) + q2 (t )σ 22 (t )

(1)

Where, the subscript 1 and 2 represents the two classes, background and foreground respectively. The estimated class probabilities and class variances are calculated as: t

q1 (t ) = ∑ P(i ),

(2)

i =1

q2 (t ) =

(t ) σ 12=

2 σ= 2 (t )

K

∑ P(i),

t

1

K

∑ [i − µ

i = t +1

iP (i ) , i =1 q1 (t ) t

µ1 (t ) = ∑

µ2 (t ) =

K

iP (i )

∑ q (t ) ,

i = t +1

2

2

(t )]2

(4)

P(i ) , q2 (t )

(5)

(7)

Where, the pixel values of an image range from 0 to K. Since, the intensities of the currency images exhibit almost uniform intensity distribution within each currency note as shown in Fig. 1, so the use of Otsu’s thresholding gives better result. C. Extraction of ROI After the image is pre-processed, the region of interest (ROI) is extracted. The ROIs for each currency notes are shown by the bounding boxes in Fig. 1, which represents the unique identification mark. The proposed algorithm extracts the ROI which comprises of the rectangular and the circular objects hence can be used as a robust feature for the detection and recognition of the specific currency. After a deep analysis of the dimensions of all of the denominations listed in Table 1, it is found that ROI exists after almost 15% from bottom-left (width wise) and 65% from the top-left (height wise) as shown in Fig. 3. By utilizing this information, the ROIs are extracted for each of the denominations. The extracting area of ROI for each note may vary in between 15% to 25% from bottom-left (width wise) and 65% to 85% from top-left. D. Displaying the Value of Currency After the extraction of ROI, the image is analyzed to check for the possible categories discussed in section 3. The circulatory property of each object in ROI is computed using (8).

Circulatory =

perimeter 2 , 4 * pi * FilledArea

(3)

P(i ) , q1 (t )

(6)

2

i = t +1

∑ [i − µ (t )] i =1

Where, µ1 (t ) and µ2 (t ) are the class means and are computed as:

Fig. 3. An example note showing width and height for extracting ROI

(8)

After experimentation it was found that this circulatory property will have values less than 1 for circular regions and would be greater than 1 for non-circular regions. The ROI is analyzed according to the following given conditions and the value associated with the specific currency is displayed. If the object is circular and only one in number then it is PKR-500, if the object is circular and two in number then it is PKR-1000, and if the object is circular and three in numbers then it is PKR-5000. Besides, if only one rectangular object is detected then it is PKR-20, if two rectangular objects are detected then it is PKR-50, and if three rectangular objects are detected then it is PKR-100. If no object is detected inside ROI then it is PKR10, because there is no special mark identity mentioned in the note of PKR-10. IV. EXPERIMENTAL RESULTS For evaluation of the proposed method, various experiments have been performed on all of the Pakistani currency denominations. Implementation of the proposed method has been carried out using MATLAB R2015a installed on windows 7, 64-bit Operating System. The results for two randomly selected currency notes of PKR-100 and PKR-5000 are shown in Fig. 4(a) and Fig. 4(b), respectively. The first row of Fig. 4 represents the original image, second row represents the thresholded image which is obtained through Otsu’s thresholding, third row shows the extracted ROIs for each of the input image, and the fourth row represents the value of the identified currency note. As can be observed that the region of interest for both of these notes have been extracted efficiently and identified accurately on the basis of categories of special identity marks that have been described in section 3. To show the effectiveness and robustness of the proposed method, the experiments are also performed on intensity varying images and the result for one randomly selected note (PKR-100) with its two variants is shown in Fig. 5.

Fig. 5. Results of currency identification on low and high intensity currency image.

Fig. 5(a) exhibits the identification result of the proposed method on a light-intensity image, whereas Fig. 5(b) represents the effectiveness of the proposed method on high-intensity image of note. As can been seen from the histogram shown in Fig. 5(a), the pixel count from gray level 0 to 50 is very low and the pixel count for larger gray level is very high, thus making this image a low intensity image. In both of these cases, the proposed method identifies the currency note accurately with high efficiency. For more clarification and understanding of the output of the proposed method, a MATLAB based GUI application has been developed that shows the overall process of the proposed method. A screenshot of the developed interface is shown in Fig. 6.

Fig. 6. GUI for currency identification Fig. 4. Results of currency identification and its associated value.

This GUI consists of the following features: loading an image of the currency note from the dataset, provide its identification status as verified if the given currency is identified accurately, displaying the corresponding currency value, conversion of the detected currency in various currencies of other countries (Chinees Yaan, Korean Wons, Dollars, etc.) and a graph that shows the relationship between the detected currency and the currency to which the detected currency is converted. The limitation of the proposed method lies in the fact that if the image is highly rotated and if the background is highly cluttered then the proposed method may lose its performance. V. CONCLUSION In this paper, a unique, efficient and simple approach for Pakistani currency recognition system has been presented. The method robustly identifies the given currency even in the presence of lighter and dark intensities on the basis of special identity marks present on the note. Experimental results reveal that the proposed method effectively recognizes the Pakistani currency of each of the seven denominations. VI. FUTURE WORK In future, this work will be extended for the detection and recognition of more currencies of different countries. Also more experiments will be performed for the currency images captured at different angles. ACKNOWLEDGEMENT The authors would like to thanks Dr. Sammer Zai for her assistance during this research. The authors also extend their gratitude to Raj Kumar Harani, Vishal Bachani, and Zara Ali Memon for their continuous support during this research. The authors would also like to extend their thanks to Mehran University of Engineering & Technology, Jamshoro, Pakistan, for providing us the resources necessary to conduct this research. REFERENCES [1] [2]

M. Sarfraz, “An Intelligent Paper Currency Recognition System,” Procedia Computer Science, Vol. 65, pp. 538-545, October 2015. J.F. Zeggeye, Y. Assabi, “Automatic Recognition and Counterfeit Detection of Ethiopian Paper Currency,” International Journal of Image, Graphics and Signal Processing, Vol.8, no. 2, pp. 28-36, Febuary 2016.

[3]

I.A. Doush, S. AL-Btoush, “Currency recognition using a smartphone: Comparison between color SIFT and gray scale SIFT algorithms,” Journal of King Saud University – Computer and Information Sciences, Vol. 29, no. 4, pp. 484 – 492, October 2017.

[4]

C. Bhurke, M. Sirdeshmukh, and M.S.Kanitkar, “Currency Recognition Using Image Processing,” International Journal of Innovative Research in Computer and Communication Engineering, Vol. 3, no. 5, May 2015.

[5]

U. Rahman, A Sargano, and U. Bajwa, “Android-Based Verification System for Banknotes,” J. Imaging, Vol. 3, no. 4, pp. 54, November 2017. T. Agasti, G. Burand, P. Wade and P. Chitra, “Fake currency detection using image processing,” 14th Conf. Series: Materials Science and Engineering, India, 2017.

[6]

[7]

G.T. Toussaint, “Proximity graphs for nearest neighbor decision rules: Recent progress,” Proceedings of the 34th Symposium on the INTERFACE, pp. 17-20, 2002.

[8]

A. Vila, N. Ferrer, J. Mantecon, D. Breton, and J.F. Garcia, “Development of a fast and non-destructive procedure original and fake euro notes,” Analytica Chimica Acta, Vol. 559, no. 2, pp. 257-263, February 2006. [9] E.H. Zhang, B. Jiang, J.H. Duan, and Z.Z. Bian, “Research on paper currency recognition by neural networks,” In Proceeding of the second international conference on machine learning and cybernetics, Vol. 4, pp. 2193-2197, 2003. [10] Q. Liu, and, L. Tang, “Study of Printing Identification Based on Multispectrum Imaging Analysis,” Proceedings of the International Conference on Computer Science and Software Engineering, Wuhan, Hubei, pp. 229-232, 2008. [11] A. Ahmadi, S. Omatu, and M. Yoshioka, “Implementing a Reliable Neuro-Classifier for Paper Currency Using PCA Algorithm,” Proceedings of the 41st SICE Annual Conference, SICE, Vol. 4, pp. 2466-2488, 2002.

[12] F. Takeda, and S. Omatu, “High Speed Paper Currency Recognition by Neural Networks,” In IEEE Transactions on Neural Networks, Vol. 6, no. 1, pp. 73-77, January 1995. [13] F. Takeda, S. Onami, T. Kadono, K. Terada, and S. Omatu, “A Paper Currency Recognition Method by a Small Size Neural Network with Optimized Masks by GA,” Proceedings of the IEEE World Congress on Computational Intelligence, Orlando, USA, Vol. 7, pp. 4243-4246, 1994. [14] E. Althafiri, M. Sarfraz, and M. Alfarras Bahraini, “Paper Currency Recognition,” Journal of Advanced Computer Science and Technology Research, Vol. 2, no. 2, pp. 104-115, 2012. [15] A.B. Sargano, M. Sarfraz, and N. Haq, “An Intelligent System for Paper Currency Recognition with Robust Features,” Journal of Intelligent and Fuzzy Systems: Applications in Engineering and Technology, Vol. 27, no. 4, pp. 1905-1913, July 2014. [16] V.K. Jain, and Dr.R Vijay, “Indian Currency Denomination Identification Using Image Processing Technique,” International Journal of Computer Science and Information Technologies, Vol. 4 no. 1, pp.126 – 128, 2013. [17] C. Liu, S. Ruan, G.Huang, Y. Jian, and L. Zhang, “Research on identification the counterfeit by recognizing the infrared images,” International Conference on Microwave and Millimeter Wave Technology, Nanjing, pp. 2081-2084, 2008. [18] M. A. Ansari, S. Zai, and Y. S. Moon, "Automatic segmentation of coronary arteries from computed tomography angiography data cloud using optimal thresholding," Optical Engineering, vol. 56, no. 1, pp. 1-8, 2017.