Abstract
We propose in this paper A Smart Transparent Neural Network Arabic Signboard Image Reader for Mobile Devices. A Transparent Neural Network (TNN), based on human behavior reading, was evaluated using more than 80 million words with high performance of time and memory consumption. The fields that we are exploring are: (1) improve the TNN to recognize Sentences (STNN) because signboards contain more than one word, (2) transform the STNN to be suitable not only for desk/lap top machines but mainly for any Smart (SSTNN) for mobile, (3) prepare the database needed for the evaluation of such smart system on Arabic signboards images taken by a smartphone, (4) evaluate the SSTNN on the new smartphone Arabic signboard images database. In this paper, we present firsthand the design steps of our SSTNN that is used as an Arabic script Android mobile application tool. Secondly, we describe our Smartphone Signboard Database (SSSD) of more than ten thousand images taken by more than 5 mobile devices of over 500 of Saudi signboards. The vocabulary of the SSSD is used for the evaluation of the SSTNN. Since our focus is on signboards in Saudi Arabia that are used to direct Arabia and non-Arabic speakers' pilgrims, a multi-language interface is provided. In fact, more than 70% of pilgrims in HAJJ are non-Arabic speakers, and many signboards used to direct pilgrims are translated only to English and sometimes to Urdu.