IEESE International Journal of Science and Technology (IJSTE), Vol . 1, No. 1, March 2012,12-17 ISSN : 2252-5297 Subject Category : Informatics 12 Fingerprint Identification System Using Wavelet Transform And Artificial Neural Network Ahsanun Naseh Khudori 1 , Muhammad Faisal 2 , Ririen Kusumawati 3 1.2.3. Information Technology Department Faculty of Science and Technology Maulana Malik Ibrahim State Islamic University Jl. Gayana No. 50 Malang Indonesia E-mail: ahsanunnaseh@yahoo.com , muhfais@yahoo.com Abstract. This research implements two methods to perform fingerprint processing. The first method is the wavelet transformation used for the fingerprint feature extraction. The second method is the back propagation artificial neural network algorithm used for the process of fingerprint identification. Fingerprint data sample is obtained from the website at http://www.bias.csr.unibo.it/fvc2004.databases.asp which can be downloaded for free. Wavelet transformation function to extract fingerprint characteristics by doing the decomposition for 4 levels. From the result of the decomposition, the coefficient having the greatest magnitude (low-frequency images) for 8x8 pixels is taken. This Characteristic is stored into the database My SQL to be inputs for back propagation artificial neural network. They are 64 input neurons. Input system is the fingerprints image and as the fingerprint identification of the owner (id, name, and address.). At the beginning of processing, the fingerprint is converted to be grayscale image. Then, it is changed to be YIQ color space, and only luminance Y as the gray factor of image is taken. To find the best combination of algorithm parameter of the artificial neural network, it is done by testing combination of parameters repeatedly. As the result, the best parameter combination with learning rate = 0.1, hidden layer neurons= 125 neurons with 15 fingerprint data is good .This parameter produces the good introduction of back propagation artificial neural network for 86.6%. From the best result of parameter combination, it is used to test the influence of the number of fingerprints toward the recognition. The result of the experiment shows that artificial neural network performance decreases along with the increasing number of fingerprint data being tested. Key-Word: Fingerprint, Wavelet, and Artificial Neural Network 1. Introduction Each person has a unique fingerprint structure. Its uniqueness is developed for biometric authentication systems than others because fingerprints have advantages such as: feasible, differ from each other (distinc), permanent, accurate, reliable and acceptable[1]. In the human’s fingerprint, a prominent part in the form of streaks or lines is called the hill, while the flat one that separates the prominent parts one another is called the valley. Figure 1. shows fingerprints, hills, and valleys on its fingerprint. Figure 1. Hill and valley of Fingerprint Wavelet is one of the image processing methods that can extract fingerprint features. With wavelet transform, the important features will not be lost when the dimensions of the image is reduced. Wavelet actually comes from the scale function. The concept of the transformation is simple. The original image is decomposed into 4 sub-new images. Each sub-image has ¼ times of the original image size. 3 sub-images spread at the upper right position, right bottom , and left bottom which is as a rough version (high-frequency image) of the original image. On the other hand, the upper right position is a soft version and it looks like the original image (low- frequency image). The grand wavelet used is haar wavelet because it is the easiest to use [2]. In contrast, the processing of fingerprint identification uses the backpropagation artificial neural network. The propagation artificial neural network is reused because it has a multilayer architecture in order to solve a complex problem[3]. Rashid, M.M et.all had researched about fingerprint verification system using neural network. Fingerprint recognition is based on the fingerprint minutiae. The result showed that the artificial neural network works well[4].