Performance Analysis of feed forward neural networks for the Recognition of Scaled and Rotated Hand Written Alphabets Using Soft-Computing Technique

ABSTRACT In this paper we are analyzing the performance of feedforward neural networks with hybrid evolutionary algorithms for the recognition of hand written English alphabets after their scaling and rotation in the test pattern set whereas in the training set straight handwritten alphabets have used. In this performance analysis the neural network is trained with hybrid evolutionary algorithm for the given training set of handwritten English alphabets and the performance of trained neural network is analyze for the test pattern set of scaled and rotated form of the patterns used in the training set. To accomplish the task we have taken two samples of each hand written English alphabets, one is considered as the training sample and another is considered as the unknown sample of the test pattern set , after applying the general mathematical algorithm for rotation and scaling for this sample along both X and Y axis. The random genetic algorithm and the hybrid evolutionary algorithm are applied to train the network for the samples of training set. Network trained with straight alphabet samples have been used to recognize the rotated and scaled samples from the set of already trained five straight alphabet’s sample. The hybrid evolutionary algorithm is taking the definite lead on the conventional approaches of neural network and soft computing techniques. The results of the experiments clearly show that the hybrid approach is efficient for the recognition of hand written samples of any shape and size most reliably.