Research On Fuzzy RBF Neural Network Applied In Face Recognition

With the development of society, the requirement of quick and effective automatic authentication for each field is becoming more and more urgent, and then how to identify a person with computer is a significative research work. As everyone knows, the biological characteristics are human being's inherent properties which have strong individual difference and self-stability, thus it is feasible to identify a people with the biological characteristics. Among biological characteristics, the use of facial feature is one of the most direct means of authentication. Comparing with other biological characteristics, facial feature has direct, friendly and convenient specialty, so it is accepted easily for users. Nowadays, research on face recognition problem is developed very fast. It contains many fields such as image process, pattern recognition and computer vision. It has two main parts: feature extraction and pattern recognition. Feature extraction extracts the feature information used for classification from facial image, and pattern recognition classifies the samples by use of the feature extracted from feature extraction. This paper continues to study along the second part.This paper studies deeply on current face recognition methods, proposes a face recognition model based on fuzzy RBF neural network, and proposes that using a hierarchical clustering method based on K-means clustering method to optimize the fuzzy RBF neural network. Firstly, this paper outlines the basic theory of face recognition and related technologies, and chooses the Independent Principal Component Analysis method for extracting facial feature after analyzing a variety of facial feature extraction methods. Secondly, this paper establishes the fuzzy RBF neural network classification model based on the research on fuzzy RBF neural network and clustering methods. Like other neural networks, fuzzy RBF neural network parameters'initial value would affect training effect, and the number of neurons in hidden layer also determines the fuzzy RBF neural network performance. According to the neural networks'common property and the fuzzy RBF neural network's characteristics, this paper proposes a hierarchical clustering method based on K-means clustering method which was used to analysis training samples, setting the fuzzy RBF neural network parameters and assigning the number of hidden layer neurons. On the basis of this optimization, this paper also proposes a method of reducing training parameters to optimize the network. According to experimental comparison, the optimization methods improve the fuzzy RBF neural network's training effectiveness and recognizing efficiency.