Optical character recognizer using artificial neural networks

Artificial Neural Net Models have been studied for many years in hope of achieving human-like performance in the fields of speech and image recognition. These models are composed of many non-linear computational elements operating in parallel and arranged in patterns reminiscent of biological neural nets. Computational elements or nodes are connected via weights that are typically adapted during use to improve performance. There has been a recent resurgence in the field of artificial neural nets caused by new topologies and algorithms, analog VLSI implementation techniques, and belief that massive parallelism is essential for high performance speech and image recognition. The pattern classification abilities of neural networks have make them suitable for practical image recognition tasks such as industrial character recognition. This paper provides the application of two important neural net models to recognition of IC characters. The aim is to ascertain the network sizes that are suitable for both rotated and unrotated characters, and the performance of these networks with untrained font types. A single method for pre-processing and representing character data was used for all networks. To limit training time, characters are considered of digits only.A significant feature in all the training sessions was the exclusion of actual IC character images in the training sets. This was to support the objective of determining the extent of font type invariance of Back Propagation (BPN) and Self-Organizing Map (SOM) networks. Lastly, it is emphasized that the results of the investigation are conclusive within the parameters of this investigation.