Novel Intelligent Pattern Recognition Methods And Its Application

As artificial intelligence theory study thoroughly, pattern recognition achieved further development and can be applied in a field and constant spread. Intelligent and pattern combined, use machine simulation human perception, including accept information, outside, so processing information, intelligence pattern recognition has become mainstream research direction, in practical application are made manifest, also, for example fingerprint identification, face recognition, etc. For improving the efficiency of intelligent pattern recognition, we do the research from two aspects including the following two parts:1. Radial basis function neural network (RBFN) is a great deal of attention and studied in recent years, that is also a member of intelligent pattern recognition method. The traditional RBFN can't meet people's requirement, so a lot of methods which combined RBFN are proposed. The combination methods can get a better result, but this practice ignores the importance of the network itself and limitations of the prototype. From a new perspective to represent the RBFN prototype, we introduce the idea of center-plane into RBFN. Euclidean distance function as the hidden node in the traditional RBFN which is the distance between the center points to the data points is replaced by the distance between the center-plane to the data points in the new RBFN. The new method makes the strong relationship between the data of data sets, especially the data itself is organized around the data center it can better expresses the data internal relations and better reflects the distribution of data and its direction. High dimensional data sets in UCI database we used to proof the effectiveness of the new method.2. In addition, the combination of different algorithm is one of the means for getting better results in the intelligent pattern recognition. In the traditional algorithm for face recognition, the most train face is positive, when the face pose changes, the rates of most recognition algorithm are greatly reduced. With the change of face pose, the original features of the face have been blocked, resulting in lower recognition rate. For solving the problem of the multi-view face, a new method is proposed which make the supervised manifold algorithm and tensor and the kernel method combination for face recognition. Through the introduction of the face information, we use the supervised algorithm to extract the view manifold from the rotation posture. In the SLPP algorithm, the neighboring points as supervised information can find the global structure as well as local structure. Combining with tensor decomposition and kernel methods, we can learn a set of nonlinear mapping functions from the embedding space into the input space to establish a model for face recognition. Experimental results show the effectiveness of the new method.