View-Based Word Recognition System

Abstract

In this paper, a new method for word recognition and classification without segmentation is presented. The worked out algorithm is based on recognizing the whole word without separating it into letters. According to this algorithm, entire words are treated and analyzed as object images subject to classification. The method is based on the "view-based approach" presented in authors' previous works for hand and printed script recognition. The top and bottom views of each word are analyzed, and the characteristic points describing it are created. The procedure of the processing and recognition involves the application of neural networks. This method is used to modify the "view-based" algorithm and improve the efficiency of the identification process. Printed words in Latin alphabet form the data base for the experiments introduced in this work. The obtained results are good and promising.