Research On Facial Expression Recognition Based On Feature Fusion

In recent years,Facial Expression Recognition(FER)has become a challenging research content in pattern recognition and computer vision.FER refers to the analysis of specific facial expression changes through computer.Let the computer understand its true inner feelings and realize practical application results such as human-computer interaction.At present,this technology is still in the research stage,and there are still many problems that need to be continuously studied by researchers to solve.In the paper first expounds the background significance of expression recognition research,the research status both inland and abroad,and analyzes the current mainstream methods at inland and abroad.The content of feature extraction in expression recognition is introduced.And analyzing the algorithms' performance.The specific research contents are as follows:Firstly,a comprehensive research on facial expression recognition algorithm is carried out,and various feature extraction algorithms are deeply analyzed.Their principle and performance are compared.The feature extraction part of this paper focuses on facial expression recognition..Secondly,a Monogenic symmetric LTP feature algorithm based on feature fusion is proposed.The characteristics of each part of the fusion are analyzed in detail,including local binary Pattern(LBP),Local Ternary Pattern(LTP),and Monogenic-signal analysis.and the improved model of these algorithms.The local binary mode and its improved version are classic and efficient texture descriptors,and because of their outstanding performance,researchers are listed in tasks such as biometrics.Monogenic signal analysis makes image feature information sufficient because of its unique feature extraction mode.Therefore,a center-symmetric local ternary encoding combining Monogenic-signal analysis(M-CSLTP)is proposed,which algorithm has stronger feature parsing ability of monogenic filtering and the stable texture representation ability of LBP.Improve the accuracy of recognition.Thirdly,for the problem of the center-symmetric local ternary pattern,a dynamic threshold adjustment method is proposed to optimize it.The problem of center-symmetric local ternary mode and monogenic filter coding matching is studied: the monogenic amplitude is encoded by the cracked central symmetric ternary mode,and the fusion expression is generated by combining the monogenic phase information and the monogenic direction information.Finally use the nearest neighbor classifier for classification.The algorithm effectively improves the recognition rate of expressions and is more robust to noise and illumination.