Knowledge Discovery Based Automated Recognition of Traffic Sign Images Using Hybrid PCA-RBF Network

Abstract

In this article, template-free, knowledge discovery, and correlation-based distance mapping have been applied for automated recognition of traffic sign images. Knowledge extraction by the neural network and correlation distance mapping in the knowledge domain have made solutions intelligent and faster. Principal Component Analysis (PCA) has been applied in the pre-processing stage to reduce the dimensionality by considering the few main components only. The proposed method has applied a self-adaptive form of radial basis function neural network to develop an efficient classifier. To increase the recognition efficiency, a firm correlation between input and neural architecture has been developed by one to one mapping approach in the learning process. The correlation-based distance measure has ensured the optimal use of discovered knowledge in the final decision stage. The developed method has shown a high level of robustness in recognition against limited visibility because of weather as well as the aging effect of traffic sign images. The proposed method can also be applied in the different applications in the area of image-based information retrieval and machine vision.