A Study of Handwritten Characters by Shape Descriptors: Doping Using the Freeman Code Chekib Gmati SITI Lab ENIT Tunis-Belvedere, Tunisia chekibgmt2007@yahoo.fr Sofiene Haboubi SITI Lab ENIT Tunis-Belvedere, Tunisia sofiene.haboubi@istmt.rnu.tn Abdulqadir Alaqeeli CRI KACST Riyadh, Saudi Arabia aalaqeeli@kacst.edu.sa Hamid Amiri SITI Lab ENIT Tunis-Belvedere, Tunisia hamidlamiri@yahoo.com Abstract! In this paper, we present the role of shape descriptors in the off-line recognition process of handwritten isolated Arabic and Latin characters. We will give some statistical and structural shape descriptors and mention their performance. Then we will present an hybrid approach that uses structural shapes’ descriptors from a different angle in order to improve the recognition’s results from a statistical descriptor. We will therefore introduce the concept of doping aiming to raise the recognition rate. Keywords-shape descriptors; statistical approaches;structural approaches; freeman code; fourier descriptor; doping I. INTRODUCTION The basic rules of school education, for learning writing and reading, explain why the handwriting was not invaded by the scan despite its scope. There is a very close relationship between the writer and handwriting. The study of this relationship allows us to know more about the personality of the writer. This justifies the usefulness of the study of handwriting in areas such as biomechanics, biometrics [1] and graphology [2] [3]. Also, the study of handwriting is the subject of several researches and its importance increases. The study of handwriting has been translated by the implementation of !handwritten characters recognition systems". The architecture of these systems comprises three main stages: the preprocessing, the feature extraction and the classification stage. The step !key" in the recognition process is the feature extraction step because it directly affects the results of the recognition system. The extraction of primitives is carried out through shape descriptors. These descriptors are the !keystone" of the handwritten character recognition system. They are divided into two categories: statistical and structural descriptors [4]. II. BACKGROUND A. Statistical descriptors The statistical methods perform statistical measures that lead to describe the shape in a comprehensive manner. Descriptors that adopt these statistical approaches are simple geometric descriptors, based on moments, Fourier descriptors and descriptors based on a transform of the image. Simple geometric descriptors are fast in computation time. They have several methods such as the degree of ellipticity or degree of squareness which is the ratio of the surface with the minimum bounding box [5]. We can mention others, like the eccentricity E and C the convexity of the object: E= Length of the major axis / Length of the minor axis (1) C= Perimeter of the convex hull / P (2) Descriptors based on moments are divided into two categories: the moments of orthogonal polynomials and moments non-orthogonal. Moments of orthogonal polynomials can be continuous or discrete. Among the continuous time in this category, we include the Zernike moments that have been proposed by [6]. This descriptor is made up of complex polynomials which form a complete orthogonal set, defined on the unit disk 1 2 2 = + y x . This descriptor is considered a descriptor robust with high descriptive power [7]. The computation time of Zernike moments is very important. Several searches were conducted in order to reduce complexity and increase the performance of Zernike moments [8] [9]. We can cite also the moments of Fourier_Mellin [10] The Pseudo-Zernike moments of generalized [11]. In the category of discrete moments of orthogonal polynomials, we quote the moments of Tchebichef [12] and Krawtchouk moments [13]. Orthogonal moments are remarkably efficient than no- orthogonal moments, a comparison between the descriptors based on moments has been reported in [14] [15].The Fourier descriptors are presented in [16] descriptors by tangent or complex representation. The complex representation is to represent the contour points of the form in the complex plane. Fourier descriptors are the coefficients of Z-transform follows: ｦ = − = N j ijk j k e z Z 1 ) 2 ( π (3) Where ݖ ௝ ൌ ݔ ௝ ൅ ݕ ௝ With ሺ ݔ ௝ ǡ ݕ ௝ ሻ the coordinates of ሼ ௝ ሽ the contour of the object, and in the complex plane  is 2012 International Conference on Frontiers in Handwriting Recognition 978-0-7695-4774-9/12 $26.00 ' 2012 IEEE DOI 10.1109/ICFHR.2012.170 285