Design And Implementation Of Facial Expression Recognition System

Facial Expression Recognition of computer technology is what the computer analyze and extract some of facial expression's features from the human facial image, according to ways of people's cognition and pattern recognition and classification methods,identify and understand the emotional.It is a challenging cross-cutting issue involved in the image processing,pattern recognition,psychology and other fields, which is a research hot spot over the past two decades,which promote feelings, human and intelligent of computers.Facial Expression Recognition is more and more important in research value and broad in application prospects.In this paper,for clues to the process of Facial Expression Recognition,I have studied correlation algorithms and progressively realized a face recognition system.For feature selection,multi-category SVM classifier,computing speed of FER in the process of existing facial expression recognition,I research several algorithms of FER,which based on the FAI,vote,AVL-tree,parallel processing theory,and the main work of the paper is as follows:1.The feature selection algorithm based on the FAI and the voting:To study the traditional attribute reduction algorithm the on FAI,I found two important flaws:a. use directly J(k)——the cumulative value of FAI as the standard reduction,which reflect the comprehensive performance of the first k-dimensional feature and is prone to bias;b.Control of reduction——"sample compatibility" on rough sets,while it is use to continuous application of eigenvalues,it is the more sensitive for discrete series,resulting time-consuming and inefficient of reduction algorithm.Improve the algorithm on the "class-pair" vote about the FAI value.The new algorithm increases the validity of attribute reduction(or feature extraction).2.The algorithms of expression classification on multi-class SVM classifier:I find structural deficiencies of a binary decision tree(or AVL-tree) based on one-to-many multi-class SVM:the non-balanced binary tree,higher a tree is,more cumulative error are caused between child class and father class,then I improve the multi-class SVM based AVL-tree.For the loophole of "one-on-one" Multi-Class SVM —the voting conflicts,the Neighbors Method is designed to resolve it.3.To explore the paralleled FER algorithms on the of multi-core:in the process of facial expression recognition,there are mass computing because of the related image processing and pattern recognition algorithm,leading to lower speed of recognition,and seriously weakening the value application of FER.On OpenMP compiler interface,the ideas and technology of Parallel processing is used to reconstructive the parallel algorithms of FER,which the high-speed processing capability can be bring into full play,and then enhance the practicality of FER.