期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2015
卷号：72
期号：3
出版社：Journal of Theoretical and Applied
摘要：This paper presents an unconstrained system for off line handwritten Amazigh character recognition based upon Legendre moments and neural networks. Legendre moments are used in features extraction phase and optimized by maximum entropy principle. Neural networks methods are chosen due to their natural flexibility and the extent of parallel processing they allow. The performance rate of our proposed scheme is strongly influenced positively upon the feature extraction algorithm that has been applied in this work. The method is robust to different styles of writing, different levels of smoothing and insensitive to the nuances of characters. Also it is performing to the translation, scale invariance. The performance of the proposed method is examined in terms of Legendre moments order and neural networks size and architecture. Features like pixel density features and Euclidean distance features are extracted from the normalized characters. The two features are applied to two different neural networks. A comparison of these techniques and our method is made. Through comparison with cited methods, the proposed method has significantly advanced and the recognition rate reaches 97.46%.
关键词：Offline character recognition; Features extraction; Legendre Moments; Neural Networks; Multilayer Perceptron
