Character Recognition Using Subpattern Coding in Neural Networks
Artificial neural networks can achieve high computation rates by using a large number of processing units with a high degree of connectivity between them. Network parameters are computed in such a way that they cause the network to converge to an equilibrium representing a solution. The aim of this paper is to give a novel biologically motivated, computationally efficient approach to character recognition using subpattern coding. Each character is decomposed into a number of different sizes of regions corresponding to subpatterns. While similarity between the character patterns considered as a whole is usually weak, it is commonly possible to obtain a great deal of similarity between their subpatterns. If similarity of subpatterns of various characters is greater than a certain level, they are assigned the same portion of character code and stored only once in a neuron. The stored subpattern together with the respective code portion is called a leaf. This similarity reduces the storage requirement for the leaf data and the testing time. During testing, the code representing a character pattern is retrieved instead of the character subpatterns which are distributed all over the leaves kept in an associative memory. Keywords: Associative memory, subpattern, leaf, similarity, correlation, stimulus.
YAZICI, Rıfat and KARAL, Hasan (1996) "Character Recognition Using Subpattern Coding in Neural Networks," Turkish Journal of Electrical Engineering and Computer Sciences: Vol. 4: No. 1, Article 2. Available at: https://journals.tubitak.gov.tr/elektrik/vol4/iss1/2