|
With in-depth theoretical study of artificial intelligence, pattern recognition research has made further development, the field can be applied also constantly open to expansion. Intelligent mode combined with the use of the machine to simulate human perception of the outside world, including to receive information, process information, etc. Therefore, intelligent pattern recognition has become the mainstream of research direction, in practical applications are also exhibited, such as fingerprint recognition, face recognition, etc. . Intelligent pattern recognition methods varied, in order to improve the efficiency of intelligent pattern recognition, this paper commenced work from two aspects, including the following two elements: 1, radial basis function neural network (Radial Basis Function Neural Network, RBFN) is In recent years, widespread attention and study neural networks, but also intelligent pattern recognition methods one, single RBFN can not meet people's needs, so RBFN combined with other methods from each other, then we can get better application effect. But this approach ignores the importance of the network itself and limitations of the prototype, ultra flat this thinking introduce RBFN, from a new perspective to interpret RBFN prototype, the traditional RBFN basis functions in the hidden nodes Euclidean distance is used function, the distance between the center point to the data. In the new RBFN, the base function takes the data center have a point distance from the hyperplane to the hyperplane data center instead of the original data center, the strong correlation between the data of the data set, in particular the data itself is centered around the data center into a distributed data plane, hyperplane can better express the intrinsic link between the data and better reflect the distribution of data and directionality. High-dimensional data sets on UCI has achieved good results. 2, in addition, a variety of algorithms is to improve the combination of intelligence pattern recognition means. In the traditional algorithms for face recognition, most front face is to estimate the face pose if changed, most algorithms recognition rate greatly decreased. With the face pose changes, the original has some facial features change with the angle is blocked, the loss of face is determining characteristic pose face recognition rate of decline led to many of the main factors. In order to solve practical process in turn affect human face gesture recognition rate problem, a supervised manifold tensor decomposition combined with face recognition algorithm. By introducing monitoring information, using supervised manifold algorithm will face rotation gesture manifolds extracted, combined with tensor decomposition and Gaussian kernel function, posture mapped into a high dimensional data space, in order to establish face recognition model. Through the high Victorian pose face test data obtained good results.
|