New Dynamic Self-Organizing Feature Maps for the classification of extracted feature vectors of characters

Abstract

The importance of neural network in pattern recognition is inevitable. Offline handwritten recognition is a major application of pattern recognition. Self-Organizing Feature Map or Kohonen map is a data visualization method which can decrease the dimensions of data by clustering the similar data. A new dynamic SOFM classification process is used in the proposed system. It can be used as a character classification process before the conversion of the handwritten image into machine readable format. The classification of input data is performed by unsupervised learning. The proposed dynamic DSOFM increases the convergence speed about 2 times as much as the speed of ordinary SOFM method. The comparison of performance analysis of DSOFM and ordinary SOFM shows that the proposed method is efficient in terms of time consumption for character classification.