The Research Of Facial Expression Synergetic Recognition Means

Facial Expression is an indispensable and important mean in the process of interpersonal communication, as it contains rich human behavior information. It has widely application prospects in human-computer interaction, psychological emotional analysis, safe driving, animation and film production fields and so on. Obviously it can bring great economic value and social benefits. So in recent years facial expression recognition technology has become currently a research hotspot.This paper first describes the application prospect and research significance of facial expression recognition. Its experiment Data is based on Japanese JAFFE expression database, and applies the theory and mean of synergetic pattern to facial expression recognition. Compared with traditional pattern recognition, synergetic pattern recognition (SPR) bases on human awareness mechanism and whole feature. It Recognizes image from top to down, so they reflect the advantage of synergetic patter recognition. Against facial expression images always contain noise information, it is geometric normalization and histogram equalization preprocessed. Then paper’s research importantly focus on two key technology of synergetic patter prototype selection algorithm and optimizing attention parameter of synergetic neural network (SNN) in the means of synergetic classification recognition. What is more, both are improved.In the synergetic pattern prototype selection algorithm, through analyzing clustering learning algorithm, this paper proposes an improved K-means algorithm. A new method for selecting initial cluster centers according to the inner class distance of samples which dynamically adjust the distance between clustering. it is easier to close between inner clustering and far away from different clustering and can effectively avoid K-means algorithm into local optimum. The algorithm training time is shorter and recognition rate is lager when clustering center as prototype pattern. To make prototype pattern has ability of self-learning, a new superposition of information learning algorithm based on adaptive parameter is proposed, the each setting of free parameter of algorithm is ratio of highest misrecognizing rate pattern whose misrecognizing samples number of the quantity of total number after training, it is not set manually but got automatically. Compared with fixed parameter, this algorithm not only can converge but faster. In the optimizing attention parameter of SNN, An improved award-penalty learning training algorithm for amending attention parameter is proposed through amending strategic and designing. The strategic is comparing misrecognizing samples rate and misrecognized samples rate. If former is lager than latter, increasing attention parameter of former corresponding to pattern, otherwise reducing that of latter corresponding to pattern. This algorithm makes training result better. Through this paper’s synergetic recognition algorithm recognize facial expression, the result of experiment shows that its classification results with this paper’ mean have been effectively improved.