Key Algorithm Research And Application For The Statistics Pattern Recognition System

The capability of pattern recognition is an important part of human intelligence. Realizing the automatic pattern recognition by computer is a key breakthrough of intelligence machine developing. Pattern recognition is not only a research field having great scientific value, but also a great key technique which we need to break through in many applications of the digital network times.Integrating the research work of some application direction, such as cell classification, speech recognitio, face detection , face recognition and alpha recognition, this paper systematic describes the composition of the statistics pattern recognition system, traces the key techniques which are widely used in many hot application, at last, lays especial emphasis on the research and creative work, and shows the application of relative technique.Feature extraction and selection, classifier design are two key techniques of pattern recognition. For the former one, we expound the technique of extracting structure feature, the technique of compressing feature space and the technique of selecting feature. For the second one, we expound multi-layers neural network and Adaboost machine learning algorithm. All of the above techniques are the popular ones of this field now.Some creative works in the paper are:Aiming at the speed problem of image pre-procession, we purposed an integral-image method for fast image processing;basing on the research of traditional face detection using Adaboost, we purposed an improved ensemblemachine learning algorithm called B-Adaboost(Backtrack-Adaboost), we could select features and construct linear dichotomizer using it, and designed a polychotomizer structure in the form of binary tree;in order to speeding up the recognition speed of polychotomizer and improving the accuracy, we purposed a new feature fusion based multi-layer classifier framework.We also research the feature space compression technique deeply.