Approaches for Offline Cursive Handwritten Character Recognition

Handwritten Character Recognition (HCR) plays an important role in the retrieval of information from pixel-based images to searchable text formats. For instance (HCR) nonlinear normalization of character size, and feature compression of higher dimensional original features are studied as pre-processing and feature extraction techniques for a statistical character classifier to improve the recognition accuracy of handwritten character recognition. A high speed pre-classification technique using a linear discriminant function is employed to improve the recognition speed. The nonlinear normalization is also utilized to supply the lack of the training samples by artificially generating character samples. The Inception V3 network is trained with character images consisting of noises which are collected from receipts and newspapers. Analysis and discussion were also made on how the different layer�s properties of neural network affects the HCR�s performance and training time. The proposed deep learning based HCR has shown better accuracy than conventional methods of HCR and has the potential to overcome in the text.