New features using fractal multi-dimensions for generalized Arabic font recognition

  • Authors:

  • Sami Ben Moussa;Abderrazak Zahour;Abdellatif Benabdelhafid;Adel M. Alimi

  • Affiliations:

  • Research Group on Intelligent Machines (REGIM), National School of Engineers of Sfax, BP W, 3038 Sfax, Tunisia and Le Havre University, Quai Frissard, BP 1137-76063, Havre, France;Le Havre University, Quai Frissard, BP 1137-76063, Havre, France;Le Havre University, Quai Frissard, BP 1137-76063, Havre, France;Research Group on Intelligent Machines (REGIM), National School of Engineers of Sfax, BP W, 3038 Sfax, Tunisia

  • Venue:
  • Pattern Recognition Letters
  • Year:
  • 2010

Quantified Score

Hi-index 0.10

Visualization

Abstract

In this work, a new method is proposed to the widely neglected problem of Arabic font recognition, it uses global texture analysis. This method is based on fractal geometry, and the feature extraction does not depend on the document contents. In our method, we take the document as an image containing some specific textures and regard font recognition as texture identification. We have combined both techniques BCD (box counting dimension) and DCD (dilation counting dimension) to obtain the main features. The first expresses texture distribution in 2-D image. The second makes possible to take on the human vision system aspect, since it makes it possible to differentiate one font from another. Both features are expressed in a parametric form; then four features were kept. Experiments are carried out by using 1000 samples of 10 typefaces (each typeface is combined with four sizes). The average recognition rates are of about 96.2% using KNN (K nearest neighbor) and 98% using RBF (radial basic function). Experimental results are also included in the robustness of the method against written size, skew, image degradation (e.g., Gaussian noise) and resolution, and compared with the existing methods. The main advantages of our method are that (1) the dimension of feature vector is very low; (2) the variation sizes of the studied blocks (which are not standardized) are robust; (3) less samples are needed to train the classifier; (4) finally and the most important, is the first attempt to apply and adapt fractal dimensions to font recognition.