Connectionist network for feature extraction and classification of English alphabetic characters

An nonadaptive connectionist architecture based feature extractor (CAFE) for English alphabetic patterns is presented. Two different adaptive connectionist networks, i.e., the multi-layer backpropagation network (MBPN) and the counter propagation network (CPN) were implemented for classification of the patterns. Their performance analysis is reported. The system is tolerant to translation and deformation and is observed to classify noisy and distorted patterns correctly.<
>