In this project, an off-line character recognition system for English alphabets based on Back Propagation (BP) neural networks has been performed. Characters were identified by analyzing their shapes and comparing their features that distinguishes each character.
The process of the system can be grouped into these stages: image acquisition, pre-processing and segmentation, feature extraction and classification based on Artificial Neural Networks using BP algorithm, finally the recognition and post- processing.
Then the System Performance is estimated and performed based on huge samples from different users and writing styles, the problems that faced the system have been discussed with some recommendations for future developments.
Visual pattern recognition such as reading characters or recognizing shapes is an easy task for human beings, but presents significant difficulty when programming an information processor to do the same thing. Artificial neural networks are a method of computation that tries to achieve human-like performance in the field of image and character recognition.
Basic BP is currently one of the most popular methods for performing the supervised learning task. The fundamental idea behind BP is that the error is propagated backward towards the minimum performance, so that a gradient search algorithm can be applied.
The goal is to adapt the parameters of the network so that it performs well for patterns from outside the training set (generalization ability).
Many reports of character recognition in English have been published but still high recognition accuracy and minimum training time of handwritten English characters using neural network is an open problem. Therefore, it is a great important to develop a handwritten character recognition system for English language. In this project, efforts have been made to develop printed and handwritten character recognition system for English language with high recognition accuracy and minimum training and classification time. Here, the following figure shows a sample that was recognized and converted to an editable text in either capital or in small letters.
In this project, an approach for recognition of printed and handwritten characters using Artificial Neural Network has been described.
Despite the computational complexity involved, artificial neural networks offer several advantages in pattern recognition and classification in the sense of emulating adaptive human intelligence to a small extent.
The number of hidden neurons was chosen depending on the input and the corresponding output and upon the experimental results, knowing that if the number of hidden nodes increases, the number of epochs taken to recognize the handwritten character is also increases.
Experimental results showed that the approach used in this project for English character recognition using BP algorithm is giving an acceptable recognition accuracy of 90% and minimum training time.