ICCCI'10 Proceedings of the Second international conference on Computational collective intelligence: technologies and applications - Volume Part III
A practical license plate recognition system for real-time environments
IWANN'05 Proceedings of the 8th international conference on Artificial Neural Networks: computational Intelligence and Bioinspired Systems
| Hi-index | 0.00 |
In this study, an artificial neural network architecture called “SimNet” is designed and implemented. SimNet is built with the following concepts in mind: simulation, simplicity, and simultaneity. It combines the general neural network structure with the subsethood concept of fuzzy logic to produce a rapid data clustering system that works somewhat similar to the well known Adaptive Resonance Theory and Self-Organizing Maps. This architecture contains both an unsupervised learning model and a supervised learning model. The former one is a 2-layer unsupervised learning paradigm, and the latter one is a 3-layer supervised learning paradigm. To demonstrate the performance of SimNet, it is applied on two applications. First, an automatic license plate recognition system (ALPReS) is designed using SimNet. In this system, 400 sample images were tested with a license plate recognition success rate of 91.25%. Moreover, the success rate of character recognition is 94.38% and the success rate of license plate identification is 98.75%. Secondly, a keypage-based data mining system using SimNet is also presented. SimNet is used in the kernel of the keypage-based data mining system for finding the similarity of the keypage and the candidate pages. The system attempts to dig out the related information from the distributed environment, the World Wide Web, according to a given webpage. SimNet has demonstrated its excellent performance in pattern recognition with the application on vehicle license plate recognition. It has also been applied on the keypage-based data mining system with promising results. Nevertheless, the preprocessing issues that occur on most of the real-life applications remain unsolved. To achieve this, domain-specific architectures with automatic preprocessing capability should be further designed for the existing neural networks, including SimNet.