New local adaptive thresholding and dynamic self-organizing feature map techniques for handwritten character recognizer

Neural network is a major tool in pattern recognition. Offline handwritten character recognizer is a significant application of pattern recognition. Binarization of the image is the major strategy in handwritten character recognition system. Various binarization procedures are analyzed here for an experimental evaluation on performance. Standard deviation from mean value is measured in the proposed system since a mean value of pixel values is not enough to find the optimal block size and the threshold value for each overlapping block of handwritten image. As an extension to the Bradley´s local thresholding, a new local adaptive thresholding is proposed. This paper deals with classification of extracted feature vectors of characters using DSOFM. A new dynamic SOFM classification process is proposed for character classification process. The proposed dynamic NDSOFM reorganizes the neural network by means of observing the furthermost and closest neurons from the neighborhood 1 of winner neuron for dynamic updating of weights. The performance analysis of NDSOFM and ordinary SOFM shows that the proposed method is efficient in terms of time consumption for character classification.