A Hybrid Neuro-Symbolic Approach for Arabic Handwritten Word Recognition

Abstract

In this article, we suggest a system that automatically constructs knowledge based artificial neural networks (KBANN) for the holistic recognition of handwritten Arabic words in limited lexicons. To build a neuro-symbolic KBANN classifier for a given vocabulary, ideal samples of its words are first submitted to a structural feature extraction module. The analysis of the presence and possible occurrence numbers for these features in the considered lexicon enables to generate a symbolic knowledge base reflecting a hierarchical classification of the words. A rules-to-network translation algorithm uses this knowledge to build a multilayer neural network. It determines precisely its architecture and initializes its connections with specific values rather than random values, as is the case in classical neural networks. This construction approach provides the network with theoretical knowledge and reduces the training stage, which remains necessary because of styles and writing conditions variability. After this empirical training stage using real examples, the network acquires a final topology, which allows it to recognize new handwritten words. The proposed method has been tested on the automated construction of neuro-symbolic classifiers for two Arabic lexicons: literal amounts and city names. The application of this approach to the recognition of handwritten words or characters in different scripts and languages is also considered.