Author(s): H.S. Al-Yousefi and S.S. Udpa
Article publication date: 1990-08-01
Vol. 8 No. 2 (yearly), pp. 49-59.
148
Keywords
Arabic characters, digitization, quadratic discriminant functions
Abstract
This paper introduces a new approach for segmenting handwritten Arabic characters to improve the recognition of these characters. The proposed approach involves, as a first step, digitization of the segmented characters. The dots and zigzags (secondaries) are then segmented and identified separately. This reduces the recognition issue from a 28 to a 20-class problem. The recognition of the primary part of the characters is achieved using features derived from the moments of the horizontal and vertical projections normalized with respect to the zero-order moment. Classification of the primary characters is accomplished using quadratic discriminant functions. Results showing considerable improvement in classification are presented.