The Artificial Neural Network Method In Face Detection And Data Mining Applications

Artificial Neural Network (ANN) is a complex network which could simulate some intelligent behaviors of human. It has been widely applied in the fields of intelligent control, system identification, and intelligent supervision etc. Based on the analysis of the model and character of ANN, this paper mainly discusses the application of it in Face Detection and Data Mining. The main work of this paper can be concluded into the following two aspects. 1.Face Detection System Based on BP Neural Network This paper designed and realized a Face Detection System, with emphasis on investigating the part of BP network whose function is to distinguish face images. BP network is a multilayer feed-forward network, which could acquire the underlying pattern of face image through supervised learning. In view of the specialty of face image, we firstly adjusted the traditional architecture of the BP network. Then, we optimized the way of tuning network's parameters, in order to avoid the defects of BP algorithm, such as inferior efficiency and easily getting into local minimum etc. Besides that, we adopted a bootstrapping method during network's training, successfully solving the deficiency of non-face training samples. Experiments demonstrate that the whole Face Detection System could identify and locate the upright, even with slight rotation, frontal faces in the given image with certain accuracy. And the optimized BP network not only learns more quickly and efficiently, but also avoids getting the local minimum to some degree. 2.Method of Clustering Categorical Data Based on GHSOM Neural Network Clustering is a very important technique and method in data mining, while the clustering of categorical data is always one of the difficult problems. Focusing on it, this paper firstly proposed a new quantization method, which could convert the categorical data, which are difficult to quantize, into high-dimensional numerical vectors. Then, an improved GHSOM network is used to cluster the quantized data. GHSOM network is a kind of neural network with unsupervised learning, good at clustering data in high dimensional space. Here, we improved the training process of GHSOM network, avoiding the original problem of granularity. Besides, we optimized the self-organizing map algorithm, improved the speed of clustering phrase.