Article Preview
Top1 Introduction
Since the beginning of the 1990s artificial neural networks (ANNs) have increasingly been used as an alternative to classic pattern classifiers and clustering techniques. Especially, during the last two decades, researchers have unravelled many aspects of the complex mechanisms underlying neural networks. A number of theories have been presented in the form of simulation models and what was really needed by end users have been developed into commercial applications. ANNs have found their way into a wide range of specific application areas, for instance: in the detection of defective objects, automatic processing of forms and mail sorting, license plate recognition, fingerprint analysis, face recognition systems, medical diagnosis of cervical cancer or breast tumours, various navigation and guidance systems or target recognition systems.
This paper covers two groups of systems. The first group was designed for automatic image recognition to recognize car license plate characters. Car license plate recognition (CLPR) systems are used in parking control systems, toll-pay systems and real-time camera monitoring. Whilst the first commercial CLPR systems appeared in 1980s, real growth occurred in the 1990s and today, although research is still continuing, there are very many systems available on the market. The second group of systems has been developed for automatic speech recognition (ASR). ASR attempts to automatically decode the speech signal by applying statistical models to minimise word error rates. Whilst ASR has already been under investigation by a large body of researches for almost four decades the results are still far from satisfactory.
One of the primary objectives of this paper is as to show how skilfully the Spatiognitron neural network (SNN) can be used to manage car license plate recognition problems in different weather and light conditions. A second objective is to show how SNN can be used to automatically recognise speech, under a variety of conditions, with the lowest possible number of recognition errors. This paper also shows key factors in achieving the best quality of results, not only during identification processes but also during other stages in the recognition process.
The organisation of this paper is as follows: sections 2.1 and 2.2 present a discussion of related work in car license plate recognition and speech recognition. Following the Spatiognitron discussion presented in Section 3; section 4 presents experimental conditions and results showing how the quality of the NeuroCar system can be improved by applying SNN. Section 5 analyses more complex recognition and presents SNN performance as applied to automatic speech recognition on a NeuroScope platform. Finally, Section 6 presents a discussion of the results, the conclusions and proposed next steps.