A Novel Feature Extraction System for Cursive Word Vocabulary Recognition using Local Features Descriptors and Gabor Filter

• Published in: 2020 3rd International Conference on Advanced Communication Technologies and Networking (CommNet) pp. 1 - 7
• Main Authors: Hamida, Soufiane, Cherradi, Bouchaib, Terrada, Oumaima, Raihani, Abdelhadi, Ouajji, Hassan, Laghmati, Sara
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.09.2020
• Subjects: Complexity theory; cursive words; Feature extraction; features extraction; Gabor filter; Gabor filters; Handwriting recognition; Histograms; KNN; pattern recognition; Urban areas
• DOI: 10.1109/CommNet49926.2020.9199642

the problems which arise in the system of automatic recognition of the handwritten Arabic script shows that the morphology complexity of the Arabic script and its cursivity remains a very vast subject of research. This research subject has known in recent years a great progress especially in the automation of postal mail sorting, check processing and form processing. Generally, there are two main types of approaches namely the analytical approach and the global approach. In this paper, we have exploited the global approach, which is based on a single description of the image of the word, seen as an indivisible entity. Our work is devoted to offline handwritten Arabic word recognition based on the global limited vocabulary approach using the IFN/ENIT database, which contains handwritten Arabic words featuring handwritten Tunisian city/village names. In the feature extraction phase, we studied three types of descriptors; the oriented gradient histogram, the local binary pattern and the Gabor filter for feature extraction. We evaluate the performance of our models by calculating some common metrics. Among the performances of three descriptors, the model based on the Gabor filter achieves an extreme precision of 99.90%.