Nonparametric Learning Using Contextual Information
A multicategory pattern classifying machine which uses contextual information is developed. In the machine proposed here recognition is based on the information of succeeding patterns as well as the features of the given pattern. It is believed that the ability to recognize is substantially improved by using contextual information, as suggested by Raviv and by Edwards and Chambers. The machine consists of piecewise linear weighting devices and maximum selectors. The learning is performed by adjusting N weighting vectors and N × N additive weighting coefficients. In this paper we show the convergence proof of the learning algorithm and present some results of computer simulations on hand-printed letters. The performance of the machine compared favorably with that of other methods.
KeywordsWeight Vector Contextual Information Weighting Coefficient Learning Procedure Convergence Proof
Unable to display preview. Download preview PDF.
- 1.R. O. Duda and H. Fossum, Pattern Classification by Iteratively Determined Linear and Piecewise Linear Discriminant Functions,IEEE Trans.EC-15(April), 220–232 (1966).Google Scholar
- 2.A. W. Edwards and R. L. Chambers, Can a priori Probabilities Help in Character Recognition?,J. ACM 11(October), 465–470 ( 1964 ).Google Scholar
- 3.I. M. Mendel and K. S. Fu, Adaptive, Learning and Pattern Recognition Systems: Theory and Applications, Academic Press, New York (1970).Google Scholar
- 4.N. J. Nilsson, Learning Machines, McGraw-Hill, New York (1965).Google Scholar
- 5.J. Raviv, Decision Making in Markov Chains Applied to the Problem of Pattern Recognition,IEEE Trans.IT-13(October), 536–551 (1967).Google Scholar
- 6.M. Shimura, Multicategory Learning Classifiers for Character Reading,IEEE Trans. SMC-3 (1973).Google Scholar