The Incremental Artificial Immune System for Arabic Handwritten Recognition

2. Register Of Names From The Tunisian National Archive

The data base used in this article is extracted from the register of names from the Tunisian national archive. The used register is composed of 32 pages where each one contains environs a hundred names. The Figure 1 shows a page of this register.

Figure 1.

Example of one page extracted from register of names

Once the page is scanned, segmentation into lines and words was applied. The extracted words constitute the database to be used for learning and classification in our system. The authors identified 234 different words with a different appearance frequency. Here, some examples of the database (see Figure 2).

Figure 2.

Some examples of the data base of names: (a) Mohammed, (b) Zouaoui, (c) Ahmed, (d) Feth, (e) Derdour, (f) Dhouadi

3. Methodology Of Work

In this article, the authors use approaches-oriented image processing and off-line handwriting recognition. First, authors prepare the input data using two methods, the first method use the NSHP-HMM (Choisy, 2006) and the second use the structural features extraction (Kacem, Aouiti, & Belaîd, 2012; Khelil, Kacem, Belaîd, & Benyettou, 2012). The authors have chosen the NSHP-HMM method because it is an elastic model used to normalize the input images to a standard size. It searches the important features and absorbs the distortions. The structural features extraction method is a classical method which gives structural properties of the image words (Choisy, 2006).