Handwritten optical character recognition by hybrid neural network training algorithm

Abstract

Handwritten optical character recognition (OCR) is the renowned research area in several fields, like writers identification, bank cheques, and so on. Literature works presented the handwritten OCR for various languages. This paper proposes a hybrid neural network training algorithm for English handwritten OCR. Initially, the noise in the input image is removed using the median filter, and the image is resized. Then, the feature sets, positional, and structural descriptors are extracted from the input image. Once the feature sets are extracted, the proposed FLM-based neural network identifies the handwritten character. The FLM proposed by combining the Firefly and the Levenberg–Marquardt (LM) algorithm for training the neural network. Finally, the proposed FLM-based neural network is integrated within the feed forward neural network, and the classification of character is done with 95% accuracy based on the size of training data, number of hidden neurons and number of hidden layers.