Support Vector Machines and Convolutional Descriptors for Arabic Handwritten Character Recognition
Abstract
The field of handwriting recognition is still a great challenge and competition between researchers; in this work, we construct a system for recognition of Arabic handwritten characters using support vector machines (SVM). Our system includes the steps: preprocessing, feature extraction, and classification. In the feature extraction phase, we simulate the behavior of convolutional neural networks (CNN) by integrating convolutional descriptors, where we used five filters: Prewitt, Sobel, Laplacian, point and line detection filter followed by a data reduction step using the Zoning method. To evaluate our approach, we created our own database of Arabic handwritten characters containing 1745 images; also, we used the HACDB database to test our system, the results we have achieved are very encouraging.