Feature extraction is the identification of appropriate measures to characterize the component images distinctly.Extracting features is one of the most important steps in any recognition system. Hence, in this paper, we explore the concept of Orthogonalized Fisher Discriminant (OFD) for unconstrained handwritten Kannada character recognition. OFD exhibits higher performance than Fisher Linear Discriminant (FLD) due to the elimination of dependences among discriminant vectors. For subsequent classification purpose, we explore the concept of probabilistic neural network (PNN) architecture. Experiments show that OFD methods are more effective and efficient than standard FLD for handwritten character recognition.

Automatic identification of a script in a given document image facilitates many important applications