Arabic recognition and translation system

Abstract

To our knowledge, there are only few systems that are able to automatically translate handwritten text images into another language, in particular, Arabic. Typically, the available systems are based on a concatenation of two systems: a Handwritten Text Recognition (HTR) system and a Machine Translation (MT) system. Roughly speaking, in the case of recognition of Arabic text images, our work has focused on the use of the embedded Bernoulli (mixture) HMMs (BHMMs), that is, embedded HMMs in which the emission probabilities are modeled with Bernoulli mixtures. In the case of Arabic text translation, our work has focused on one of the state-of-theart phrase-based log-linear translation models. In this work we evaluate our system on the LDC corpus introduced in the NIST OpenHaRT 2010 and 2013 evaluations. Very competitive and promising results are shown. Additionally, we present the idea of a simple mobile application system for image translation that recognizes the Arabic text in an image and translates the recognized text into English.