THE SCIENCE AND INFORMATION ORGANIZATION www.thesai.org | [email protected] IJACSA Special Issue on Selected Papers from Third international symposium on Automatic Amazigh processing (SITACAM’ 13) Associate Editors Dr. Zuqing Zhu Service Provider Technology Group of Cisco Systems, San Jose Domain of Research: Research and development of wideband access routers for hybrid fibre-coaxial (HFC) cable networks and passive optical networks (PON) Dr. Ka Lok Man Department of Computer Science and Software Engineering at the Xi'an Jiaotong- Liverpool University, China Domain of Research: Design, analysis and tools for integrated circuits and systems; formal methods; process algebras; real-time, hybrid systems and physical cyber systems; communication and wireless sensor networks. Dr. Sasan Adibi Technical Staff Member of Advanced Research, Research In Motion (RIM), Canada Domain of Research: Security of wireless systems, Quality of Service (QoS), Ad-Hoc Networks, e-Health and m-Health (Mobile Health) Dr. Sikha Bagui Associate Professor in the Department of Computer Science at the University of West Florida, Domain of Research: Database and Data Mining. Dr. T. V. Prasad Dean, Lingaya's University, India Domain of Research: Bioinformatics, Natural Language Processing, Image Processing, Expert Systems, Robotics Dr. Bremananth R Research Fellow, Nanyang Technological University, Singapore Domain of Research: Acoustic Holography, Pattern Recognition, Computer Vision, Image Processing, Biometrics, Multimedia and Soft Computing (i) www.ijacsa.thesai.org IJACSA Special Issue on Selected Papers from Third international symposium on Automatic Amazigh processing (SITACAM’ 13) CONTENTS Paper 1: Application of Data Mining Tools for Recognition of Tifinagh Characters Authors: M. OUJAOURA, R. EL AYACHI, O. BENCHAREF, Y. CHIHAB, B. JARMOUNI PAGE 1 – 4 Paper 2: Annotation and research of pedagogical documents in a platform of e-learning based on Semantic Web Authors: S. BOUKIL, C. DAOUI, B. BOUIKHALENE, M.FAKIR PAGE 5 – 10 Paper 3: Hierarchical Algorithm for Hidden Markov Model Authors: SANAA CHAFIK, DAOUI CHERKI PAGE 11 – 14 Paper 4: Review of Color Image Segmentation Authors: Abderrahmane ELBALAOUI, M.FAKIR, N.IDRISSI, A.MARBOHA PAGE 15 – 21 Paper 5: Invariant Descriptors and Classifiers Combination for Recognition of Isolated Printed Tifinagh Characters Authors: M. OUJAOURA, R. EL AYACHI, B. MINAOUI, M. FAKIR and B. BOUIKHALENE, O. BENCHAREF PAGE 22 – 28 Paper 6: Handwritten Tifinagh Text Recognition Using Fuzzy K-NN and Bi-gram Language Model Authors: Said Gounane, Mohammad Fakir, Belaid Bouikhalen PAGE 29 – 32 Paper 7: Performance evaluation of ad hoc routing protocols in VANETs Authors: Mohammed ERRITALI, Bouabid El Ouahidi PAGE 33 – 40 Paper 8: Recognition of Amazigh characters using SURF & GIST descriptors Authors: H. Moudni, M. Er-rouidi, M. Oujaoura, O. Bencharef PAGE 41 – 44 Paper 9: Printed Arabic Character Classification Using Cadre of Level Feature Extraction Technique Authors: S.Nouri, M.Fakir PAGE 45 – 48 (ii) www.ijacsa.thesai.org IJACSA Special Issue on Selected Papers from Third international symposium on Automatic Amazigh processing (SITACAM’ 13) Application of Data Mining Tools for Recognition of Tifinagh Characters M. OUJAOURA, R. EL AYACHI O. BENCHAREF, Y. CHIHAB B. JARMOUNI Computer Science Department Computer Science Department Computer Science Department Faculty of Science and Technology Higher School of Technology Faculty of Science Sultan Moulay Slimane University Cadi Ayyad University Mohamed V University Béni Mellal, Morocco Essaouira, Morocco Rabat, Morocco. Abstract—The majority of Tifinagh OCR presented in the literature does not exceed the scope of simulation software such as Matlab. In this work, the objective is to compare the classification data mining tool for Tifinagh character recognition. This comparison is performed in a working environment using an Oracle database and Oracle Data Mining tools (ODM) to determine the algorithms that gives the best Recognition rates (rate / time). Keywords—OCR; Data Mining; Classification; Recognition; Fig. 1. Tifinagh characters – IRCAM. Tifinagh; geodesic descriptors; Zernike Moments; CART; AdaBoost; KNN; SVM; RNA; ANFIS The Tifinagh alphabet has several characters that can be obtained from others by a simple rotation, which makes I. INTRODUCTION invariant descriptors commonly used less effective. For this The Optical Character Recognition (OCR) is a rapidly reason, we used a combination of Geodesic descriptors [5] and expanding field in several areas where the text is the working Zernike moments [6]. basis. In general, a character recognition system consists of several phases [1, 2, 3, 4, 8]. The extraction is a phase that A. Geodesic descriptors focuses on the release of attributes from an image. In this A geodesic descriptor is the shortest path between two article, the geodesic descriptors and Zernike moments are two points along the spatial deformation of the surface. In the case approaches used to calculate the parameters. The effectiveness of binary images; we used a Shumfer simplification which of the system is based on the results given by the classification comprises those operations: phase using data mining tools, which is the purpose of this document. Calculate the number of pixels traveled between the two points; The rest of the paper is organized as follows. The Section 2 Penalize the transition between horizontal and vertical pixels discusses the first primordial task of any recognition system. by 1 and moving diagonally by 1.5; It’s the features extraction problems in addition to a brief formulation for geodesic descriptors and Zernike moments as Choose the optimal path. features extraction methods. The Section 3 is reserved for the We consider the preliminary processing that consists of two second important task which is the classification problems standard processes: (i) the noise elimination and (ii) the using some classifiers based on several algorithms like ANFIS, extremities detection (Fig. 2). ANN, SVM, CART, KNN and AdaBoost. Finally, the Section 1) Extremities detection 4 presents the experimental results for the recognition system. In order to identify the extremities, we use an algorithm II. FEATURES EXTRACTION that runs through the character contour and detects the nearest points to the corners of the image. Tifinagh is the set of alphabets used by the Amazigh population. The Royal Institute of Amazigh Culture (IRCAM) has normalized the Tifinagh alphabet of thirty-three characters as shown in Fig. 1. 1 | P a g e www.ijacsa.thesai.org IJACSA Special Issue on Selected Papers from Third international symposium on Automatic Amazigh processing (SITACAM’ 13) inside the unit circle as x2 + y2 1. The form of such (a) (b) polynomials is [7]: * Vnmx, y Vnm, Rnm().exp( jm) Where: n: a positive or null integer; (d) (c) m: an integer such that | m | n ; r: length of the vector from the origin to the pixel (x, y); Fig. 2. Example of the extremities of three Tifinagh characters. θ: angle between the vector x and p; 2) Geodesic descriptors calculation Rnm: radial polynomial. We called "geodesic descriptors" the distances between the four detected extremities of the character divided by their V * (x, y): complex polynomial projection of f (x, y) on the Euclidean distances. We set: space of complex polynomials. Such polynomials are orthogonal since: DlM(xy): Geodesic distance between x and y; dxy: Euclidean distance between x and y; V * x, y .V (x, y)dxdy 0 or 1 a, b, c and d: Detected extremities of each character. nm pq And, we call: x2 y2 1 st 1 metric descriptor: D1 DlM ab da b The geometrical Zernike moments are the projection of the 2nd metric descriptor: D Dl ac d function f (x, y) describing an image on a space of orthogonal 2 M a c polynomials generated by: rd 3 metric descriptor: D3 DlM ad da d 4th metric descriptor: D Dl bc d n 1 * 4 M b c Anm f (x, y).Vnm(,)dxdy th x y 5 metric descriptor: D5 DlM bd db d th 6 metric descriptor: D6 DlM cd dc d For identification of the image, the Zernike moments modules are used: To ensure resistance to scale change of the proposed descriptors, we divided each geodesic path by the corresponding Euclidean distances. 2 2 Anm Re (Anm) Im (Anm) III. CLASSIFICATION The choice of the classifier is primordial. It is the decision element in a pattern recognition system. In this context, we compared the performance of six Data Mining algorithms. A. CART Algorithm CART (Classification And Regression Tree) builds a strictly binary decision tree with exactly two branches for each decision node. The algorithm partitions or divides recursively the training set using the principle of "divide and conquer" [9]. B. KNN (k-nearest neighbor) Fig. 3. Example of Geodesic descriptors calculation. The k-nearest neighbors algorithm (kNN) [10] is a learning B. Zernike moments method based instances. To estimate the associated output with a new input x, the method of k nearest neighbors is taken into The Zernike moments are a series of calculations that account (with the same way) the k training samples whose converts an image into vectors with real components entrance is nearest to the new input x, according to distance representing moments A . ij measurement to be defined. By definition, the geometrical moments of a function f(x, y) C. SVM (Support Vector Machines) is the projection of this function on the space of polynomials generated by xp yq where (p, q)N2. Zernike introduced a set Support Vector Machines (SVM) [11] are a class of of complex polynomials which form an orthonormal basis learning algorithms that can be applied to any problem that 2 | P a g e www.ijacsa.thesai.org IJACSA Special Issue on Selected Papers from Third international symposium on Automatic Amazigh processing (SITACAM’ 13) involves a phenomenon f that produces output y=f(x) from a TABLE I. OBTAINED RESULTS: RCOGNITION RATES AND EXECUTION set of input x and wherein the goal is to find f from the TIME observation of a number of couples input/output.