Isolated character recognition based on sub-character primitive features

Abstract

In this chapter character recognition is tried with another approach based on the sub-character primitive features. Novel features have been proposed and experimented. In this approach, on-line data is converted into series of x, y coordinates first. Some preprocessing is applied to the data and then filtered data is encoded into direction vector. A proposed smoothing algorithm smoothes the encoded sequence of direction vectors before extracting the features. These features are then fed to the input of BPN and CPN for training and recognition purposes. An overview of the whole system is presented in Figure 1. The flow of data during training is shown by the dashed line arrows, while the data flow during recognition is shown by solid line arrows. In the following sections, the techniques and algorithms to be involved for each processing block are briefly introduced.