Crucial combinations for the recognition of handwritten letters

Crucial features are important for the identification of patterns. Great ef forts have been devoted to discover the most distinctive features. This pap er presents the ideas of splitting patterns into 4- or 6-parts, integration for recognition, and selection of crucial combinations. The crucial combin ations are explored with particular reference to 26 handprinted English let ters. This paper proposes the first simple algorithms to perform the crucia l combinations. Also, the largest confusion regions and algorithms for Endi ng them are provided. The most useful crucial combinations and the largest confusion combinations are listed in this paper. Moreover, we extend the cr ucial combinations and largest confusion regions for the threshold recognit ion rates rho(ij) greater than or equal to rho(0), say rho(0) = 100%, 80% a nd 50%. The second algorithms are developed, and numerical results for 4-pa rtitions are provided. (C) 2000 Elsevier Science B.V. All rights reserved.