A Comparison of the Self-Organizing Map and Growing Neural Gas Network in the Context of Optical Character Recognition

With the advent of tablet and touch screen computing, the number of applications that utilize written text recognition technologies is rapidly increasing. These applications rely on fast and accurate optical character recognition (OCR) algorithms. The selforganizing map (SOM) is an unsupervised algorithm capable of great performance in machine learning. As such, it has become a benchmark in OCR. In spite of its popularity, it does have some significant shortcomings. Its training algorithm is inefficient, its accuracy is highly dependent on a classification algorithm, and it generates unnecessary hybrid output nodes. This research investigated alternative data structures for application in optical character recognition. The growing neural gas network (GNG) is an alteration of the SOM that was developed to address these shortcomings. The purpose of this research was to assess whether the GNG network compensates for the SOM’s shortcomings in the context of OCR.