A novel approach for Persian/Arabic intelligent word recognition

Abstract

In this paper we present a novel approach for offline Persian/Arabic intelligent word recognition based on the fast and customized dynamic time warping method. The main focus of paper is on Persian language but considering the common character sets and writing styles in both Persian and Arabic, our system could be easily extended to Arabic language. Recent advances in this area show that many systems for intelligent word recognition use either Neural Network or Hidden Markov Model that suffer from low recognition rate, sensitivity to noises or wide range of parameters that reduce system performance. The experimental results are provided by using a benchmark dataset of Persian handwritten words of 380 individual writers and it shows the proposed algorithm has the recognition rate above 90%.