Abstract
The improvement of automated systems that are capable of recognizing human handwritings offers a new way of improving human-computer interface and of enabling computers to process handwritten documents more efficiently. In order to implement a high accuracy approach for handwriting recognition system, we need several pre-processing steps for the purpose of preparing the data for the feature extraction stage. There are many pre-processing steps used in this approach to enhance the perspective of our dataset, such as noise removal, image normalization and skeletonization. Two types of feature extraction techniques have been applied in this approach statistical and geometrical. Our experiments show that the statistical feature is reliable, accessible and offers more accurate results. An adaptive multi-layer feed-forward back-propagation neural network was used in our research. This study focuses only on Malay handwritten cheque words recognition using lexical matching on top of character recognition. The presented results show that our approach has successfully increased the accuracy of recognition from 98.15% by using pure character recognition to over 99% by using the newly proposed hybrid character and lexical cheque words recognition.