International Journal of Excellence Innovation and Development ||Volume 1, Issue 1, Nov. 2018||Page No. 046-049||
Artificial Neural Network Based Offline Signature Recognition System Using Local Texture Features Shivashankar M. Rampur Professor, Department of Computer Science and Engg., Brahmdevdada Mane Institute of Technology, Solapur Belat Tal.North Solapur, Distt. Solapur, Maharashtra, India
Abstract––Signature of an individual is a significant biometric attribute of a human being which can be used to certify human identity and these attributes can have own identity like face recognition, fingerprint detection, iris inspection and retina scanning. In this work, we are developing method which deals with the off-line signature recognition system using artificial neural network in which the signature is captured and presented in the form of an image to the system. Offline signature recognition system is a significant biometric technique, which is used to offers automated process of recognition and verification by extracting local features that classifies each input signature and has many number of uses. The proposed system has local texture features and feed forward back propagation in an artificial neural network classifier to identify and authenticate signatures of individuals. Various image processing techniques are used to categorize and validate the signature. Index Terms––Image acquisition, RGB-to-Grayscale conversion, feature extraction, artificial neural network
INTRODUCTION Signature is generally acknowledged & used as a way for authorization in our daily life, which is an important biometric attributes of human used to verify human identification. Manual signature is fundamental procedure for individual, which is used for uncovering of the document of signer with the assumption that the signature varies slowly & virtually unfeasible to falsify without detection. A signature is difficult to replicate and broadly used to identify an individual delivering his day by day events such as document study, bank activities, electronic funds transfer and access control. A signature as a behavioral biometric encrypts the ballistic actions of an individual and allows higher intra-class and time inconsistency, estimates the physical qualities which are fingerprint, iris or face. Depending on exhaustion, psychological and physical state, and lettering location (ergonomics), signatures vary. The marker accelerations, which are comparative to the muscle forces exerted by the signer, are reliable in a usual signature.
local features that classifies each input signature based on artificial neural network and has many number of uses. The neural networks is the most outstanding way of finding solution of the problems that are most difficult to solve by traditional computational methods. The advantage of neural network is no need to understand the solution. While signer is signing, there are variations in terms of pen width, additions found in strokes, exchange or qualified point of strokes, scaling within the genuine signatures and rotation. Our system is motivated to overcome these variations. Our system gives high level of accuracy. Objectives The objective of our system is used to develop preprocessing phase which is processed on input signature image. This preprocessing phase include conversion of original image into grey scale, conversion of grey into binary image, noise reduction, thinning and resize. The objective of our system is used to develop feature extraction phase for classifying signature. In this phase, we are extracting texture features from signature which are entropy, homogeneity, contrast, correlation and energy. The objective of our systems is used to recognition of signature by signers. In this phase, we have to compare the texture features of test images with features of train images. If it’s matched then the given signature is identified else not.
OVERVIEW OF SYSTEM Offline signature recognition system is an automated process of detection by extracting local features that classify each input signature. In this system, we initiate by images are scanned using scanner, elaborated the input signature by preprocessing, the extraction of texture features from the preprocessed images and analyze the signature with the signature stored in the knowledge base using classification technique. If its match then input signature is recognized else not.
Motivation Offline signature recognition system is a significant biometric technique, which is used to offers automated process of recognition and authentication by extracting www.ijeid.com
{IJEID © 2018} All Rights Reserved
Fig. 1 System overview. Page | 46
Artificial neural network based offline signature recognition system
PROPOSED METHODOLOGY The proposed methodology consists of several essential steps which are supposed to be followed with précised rules. The block diagram of our proposed system gives the short idea about the procedure of methodology.
Rampur S.M.
and unwanted elements. To overcome the corruption caused due to noise, we use median filter for smoothing and recovering images. Median filter is non-linear operation which is used to reduce noise and preserve edges. This conversion is shown in Figure 4. Thinning It is one of the morphological procedures, which is constructed with procedures on pixel sets. Morphological operations take two arguments: binary image and structuring element. To conserve the aspect ratio of signature image, thinned signature image goes through normalization step. The thinned image consists of 0’s and 1’s constituted by pixel sets, where the pixels in the signature become less. This procedure is shown in Figure 5.
Fig. 2: Block diagram of proposed work. The major steps of the signature recognition are explained as follows: Image Acquisition The action of retrieving an image from the hardware based source is known as image acquisition, so that it can pass through the processes which occur after image acquisition. The first step in the workflow sequence is image acquisition, because without an image, processing is impossible. Handwritten signatures are signed on the white paper by the individuals at different timing under different emotions such as stress and joy levels and these signatures are scanned by using scanner of desired dpi resolution in jpeg image format for the creation of the database. Pre-Processing It is a technique to augment raw images by removing distortions and is the first part of the proposed system prior to computational processing. We have used RGBto-grayscale conversion, gray scale to binary image conversion, noise reduction, thinning and resizing for pre-processing. RGB to Gray-Scale Conversion In this phase, if the original image is in the form of red, green and blue which is used to convert to gray scale by using following formula: Gray-scale = (0.299*R) + (0.5876*G) + (0.114*B) It is a general performance in the image preprocessing phase, since processing of a three channel signature (colored image) is slower than that of processing a single channel signature (gray-scale image). This conversion is shown in Figure 3.
(a)
(a)
(b)
Fig. 4: Noise reduction of image. (a) image contains noise. (b) noise reduction.
(a)
(b)
Fig. 5: Thinning of image. (a) binary form of image. (b) thinned image. The Feature Extraction Phase It is the most significant phase in any recognition system because accuracy of the recognition completely depends on the features which are going to be extracted from the signatures. The main purpose of the feature extraction method is used to get back the features accurately. Here, we have to extract local texture features like entropy, correlation, contrast, homogeneity and energy.
Noise Reduction A signature images are corrupted due to addition of unwanted elements or noisy channels and also gets ruined due to destructive effects caused from lighting www.ijeid.com
(b)
Fig. 3: RGB form of image to gray-scale conversion. (a) RGB form of image. (b) Gray-scale form of image.
{IJEID © 2018} All Rights Reserved
Fig. 6: Horizontal division. Page | 47
International Journal of Excellence Innovation and Development ||Volume 1, Issue 1, Nov. 2018||Page No. 046-049|| The network collects neurons from the input level and the output of the network is displayed on an output layer.
RESULTS AND DISCUSSION
Fig. 7: Block Image divisions. Classification This is the step where we obtain approximate or appropriate result. Classification will be executed on the basis of features extracted in the feature space. Classification divides the feature into different classes based on decision rules. Signature stored in the Knowledge Base is subjected to classification technique. We have used Artificial Neural network as classifier in our signature recognition system. In classification we compare the test images with trained data images in order to classify test images.
The experiment has been approved out so as to estimate calculated system’s performance. The experiment has been worked on the database which consists of classes of 95 persons and 10 signatures per individual class where 80 textual features were extracted from each block of the signature image. The textual features are listed as: Energy Correlation Entropy Homogeneity Contrast Identification rate of this experiment is 85-90% which satisfied our requirement. The performance was checked against number of person signature also. Initially we started with database having 20 persons. Gradually we increased our database by 20 persons in each step. The following table 1 shows that 20 persons with accuracy. Table 1: Signatures with accuracy. Persons signature Accuracy 20 99% 40 98.3% 60 97% 80 96% 95 95%
Fig. 8: Artificial neural network. The computational model such as artificial neural networks is stimulated by the brain, which are capable of recognizing a pattern and machine learning. The artificial neural networks are generally organized as interrelated "neurons" and by passing information through the network the values from inputs can be determined. In general, neural networks are organized into three layers such as input, hidden and output layer. An 'activation function' is held by interconnected ‘nodes’ which are included in neural network. By means of the 'input layer', patterns are given to the network, which correspond to one or more 'hidden layers’, by using a system of subjective ‘connections’ the actual processing is done and then the hidden layers connect to an 'output layer' which gives output of the system shown in the figure 8 above. Many artificial neurons together compose an artificial neural network and are correlated according to accurate network design. The goal of the neural network is to convert the inputs into significant outputs. Here, we have used an algorithm called feed forward back propagation in ANN for identification and confirmation of signatures of individuals, where the simulated neurons are structured into layers and transmit their signals “forward” and the errors are transmitted “backward”. www.ijeid.com
We have checked performance of 20 signatures of 20 persons; keep adding by 20 persons in each step. We have trained 20 signatures per each person using feed forward back propagation algorithm. The performance rate has checked against signature as shown in following charts:
Performance Rate 100 80 60 40
Performance Rate
20 0
Fig. 9: Performance rate. The above graph shows as a number of individual increases, performance getting decreases.
CONCLUSION Offline signature recognition system is a significant biometric technique, which is used to offers automated
{IJEID © 2018} All Rights Reserved
Page | 48
Artificial neural network based offline signature recognition system process of recognition and authentication by extracting local features that classifies each input signature and has many number of uses. The proposed system has local texture features and feed forward back propagation in an artificial neural network classifier to identify signature. By using training algorithm called feed forward back propagation, we can train a neural network using extracted features from the signatures. The artificial neural network has given expected results. Our proposed system exhibited 85-90%success rate by identifying the signatures correctly that it was taught for purpose. Few more efficient features can be included to improve our system. The performance of the system can be increased by using additional features in dataset. We can increase accuracy of identification by increasing signature of trained images for each individual. Our system gives 90-99% accuracy based on number of input signature.
REFERENCES [1]
[2]
[3]
[4]
[5]
J.P. Drouhard, R. Sabourin, M. Godbout, “Evaluation of a Training Method and of Various Rejection Criteria for a Neural Network Classifier Used for Off-Line Signature Verification”, IEEE Int’l Conf. Neural Networks, Orlando, Fla., June 26-July 2, pp. 294-4,299, 1994. R. Baeza-Yates, G. Valiente, “An Image Similarity Measure Based on Graph Matching”, IEEE University of Chile, 2000. E. J. R. Justino, F. Bortolozzi, and R. Sabourin, “Off-Line Signature Verification using HMM for Random Simple and Skilled Forgeries”, Proc. 6th Intl. Conf. On Document Analysis and Recognition, 2001, pp. 450-453. Xiao, X. and Leedham, G (2002), “Signature Verification using a Modified Bayesian Network. Pattern Recognition”, 2002, vol. 35, no. 5, pp. 983-995. B. Fang, C.H. Leung, Y. Y. Tang, K. W. Tse, P. C. K. Kwok and Y. K. Wong, “Off-Line Signature
www.ijeid.com
Rampur S.M.
Verification by the Tracking of Feature and Stroke Positions,” Pattern Recognition, vol. 36, pp. 91– 101, 2003. [6] Vamsi Krishna Madasu, “An Automatic Offline Signature Verification and Forgery Detection System”, University of Canberra, Pattern Recognition Technologies and Applications: Recent Advances, 63-89, 2004. [7] M. Hanmandlu, M. H. M. Yusof, and V. K. Madasu, "Off-Line Signature Verification and Forgery Detection using Fuzzy Modelling”, March-2005, vol. 38, pp. 341-356. [8] Johannes Coetzer, “Off-Line Signature Verification”, Journal, University of Stellenbosch, April -2005, 45-90. [9] Juan J. Igarza , Inmaculada Hernáez, Iñaki Goirizelaia, Koldo Espinosa and Jon Escolar, “Off-Line Signature Recognition based on Dynamic Methods”, Dept. of Electronics and Telecommunications, University of the Basque Country Alameda Urquijo s/n, Bilbao, Spain E48013, vol.5779,2005. [10] Dakshina Ranjan Kisku, Phalguni Gupta and Jamuna Kanta Sing, “Fusion of Multiple Matchers using SVM for Offline Signature Identification”, Communications in Computer and Information Science, 2009, Volume 58, pp. 201-208.
Biography of Author
Mr. Shivashankar M. Rampur received the B.E and M.Tech degrees in Computer Science and Engineering from Basaveshwar Engineering College, Bagalkot, Karnataka in 2012 and 2016, respectively. Presently working as Assistant professor at Brahmdevdada Mane Institute of Technology, Solapur.
{IJEID © 2018} All Rights Reserved
Page | 49