Arabic text detection using ensemble machine learning

Article type: Research Article

Authors: Oulladji, Latifa^{a; *} | Feraoun, Kamel^a | Batouche, Mohamed^b | Abraham, Ajith^c

Affiliations: [a] Djillali Liabes University, Sidi Bel Abbes, Algeria | [b] Constantine 2 University, Constantine, Algeria | [c] Machine Intelligence Research Labs (MIR Labs), Scientific Network for Innovation and Research Excellence, WA, USA

Correspondence: [*] Corresponding author: Latifa Oulladji, Djillali Liabes University, Sidi Bel Abbes, Algeria. E-mail: [email protected].

Abstract: The automatic detection and recognition of zone text in natural images remain indispensable due to the omnipresent of text information in daily human life. This domain contoured a development of many applications specially with English language where many systems were implemented and proved their efficiency. Arabic language represents a real challenge for its cursive nature and rich vocabulary. The first step of our work was inspired from Gomez and Karatzas [7] on multiscript detection using Gestalt theory. For the second step, we implemented three classifiers namely Neural Network (NN) Support Vector machine (SVM) and Adaboost. These classifiers were deployed to classify the group regions in images as text or non-text. To improve the system performance an ensemble method based on majority voting was applied where the outputs of the three classifiers were fused. Experiments were conducted using own image database and ground-truth and the empirical results illustrate that the proposed method is efficient.

Keywords: Arabic text detection, gestalt theory, neural network, adaboost, support vector machine

DOI: 10.3233/HIS-180254

Journal: International Journal of Hybrid Intelligent Systems, vol. 14, no. 4, pp. 233-238, 2018