Research On Facial Expression Recognition Based On Rough Set And Fused Feature

The Face Expression Recognition (FER) refers to use the computer technology to extract features from the face expression iimage according to the human thinking way to classify the features, in order to analyze and understand the human feelings from the human face information. The FER is a challenging cross-cutting issue which is related to biological features recognition, pattern recognition, image processing, machine vision, movement tracking, physiology, psychology and so on. At the same time, it is an important part of affective computing and intelligent human-computer interaction which has broad applications and potential market values.In this thesis, we firstly discuss the background and then analyze the main exiting expression feature extraction and recognition algorithms. On this basis, we propose the method of Facial expression Recognition Based on Rough Set and Fused Feature. The main contents are described as bellows:(1) The rough set is introduced to FER, and the algorithm of improved rough set attribute reduction is proposed, it makes up the shortcomings of the original attribute reduction algorithm which only embarks from the kernel attributes. It considers the importance degree of the attribute takes into account the contributions degree of various attributes. So it can be used to select expression features effectively.(2) To overcome the problem of existing WPCA which only stressed some or two region information and weighted regions shape is not flexible enough, the updated weight principal components analysis algorithm (UWPCA) is proposed. It scatters emphasis on the regions which have important contributions to the expression changes such as the eye, eyebrow and mouth, so the face expression features are more prominent. Then, two coefficients of the horizontal and vertical direction are joined to achieve two-way adjustable of the weight region.(3) In order to obtain more effective feature, a new method based on the partial features and the overall features is proposed to extract the hybrid features. AAM method is used to partial position, select 32 key points from the eyebrows, eyes, nose, cheek and mouth areas, calculate the different distances between different points, Use the rough set to select the featuresr to achieve the partial geometry features; UWPCA is used to the overall features from the expression images with the rough set. Then the KCCA is used to fuse these two features as the observation vectors of discrete HMM, which can achieve a good result.(4) A prototype system of facial expression recognition based on rough set and fused feature is designed and implemented. The modules in this system mainly include: Image pretreatment, expression feature extraction, feature selection, emotion feature fusion and typical expression recognition.