Neural Network for the Recognition of Handwritten Tunisian City Names

Abstract

The complexity of the Arabic characters morphology makes research in recognition of the handwritten Arabic writing remain an interesting topic. In this setting, a system for recognition of handwritten Arabic words based on a Transparent Neural Network, called TNN-DF is developed within the LSTS laboratory. It uses structural features to describe words and makes recourse to Fourier descriptors (DF) when encounters an ambiguity. To enhance recognition results of TNN-DF, we suggest a neural approach to learn letters, part ofarabic words and words. Experiments conducted on 750 samples, of 50 city names, extracted from the standard IFN/ENIT' database of handwritten Tunisian city names show an improvement of recognition accuracy. The results are promising, and suggestions for improvements leading to recognition of larger voca bulary are proposed.