Research On Facial Expression Recognition Based On Local Feature

In order to make intelligent devices interact with humans more humanized,it can be realized by recognizing human facial expressions and perceiving emotional states.Therefore,facial expression recognition has great scientific research value and broad application prospects.In recent years,a lot of research has been done on Image-based facial expression recognition,but fast and accurate facial expression recognition is still a daunting task due to factors such as occlusion,illumination,and pose.Aiming at the problems existing in some current local feature extraction algorithms,the main research contents of this paper are as follows:(1)The Local Direction Pattern(LDP)and Local Direction Number Pattern(LDN)are analyzed.The LDP pattern ignores the direction information in the gradient space,and LDP and LDN pattern ignore the gray information,Local Prominent Direction Texture Pattern(LPDTP)is designed,Firstly,the prominent directional features are found in the gradient space of the image for coding,Then in the intensity space of the image,the gray information in the prominent direction is encoded,the coding features with gradient information and gray information are obtained;In the flat area of the face,due to the small amount of information,set an automatic threshold θ,The coding features of the flat region are removed according to the gradient intensity.Experiments show that the feature representation ability and robustness to gaussian noise of LPDTP are better than LDP and LDN.(2)Considering that Local Binary Pattern(LBP)can effectively encode image texture information,but the feature dimension is high.By analyzing the Central Symmetric Local Binary Pattern(CS-LBP)and the Central Symmetric Local Derivative Pattern(CS-LDP).the CS-LDP is improved,and the Multi-directional Central Symmetric Local Binary Pattern is proposed(MCS-LDP),By using four feature descriptors to encode the image texture features from different directions,and then cascade fusion the extracted feature vectors,the image texture information can be fully and effectively extracted.Principal Component Analysis(PCA)is introduced to reduce the dimension of the fused features,remove the redundant information and reduce the feature dimension.Experiments show that the recognition rate of MCS-LDP is significantly improved compared to CS-LBP and CS-LDP.(3)Due to the limited representation ability of a single feature,image information can be more effectively represented by fusing multiple features.The LPDTP proposed in this paper mainly extracts the gradient information of the image,and the features are stable;the proposed MCS-LDP mainly extracts the texture information of the image,and the feature recognition rate is high.The feature vectors of these two algorithms can complement each other well.The LPDTP feature vector and the MCS-LDP feature vector are fused at the feature layer using the serial fusion method,and PCA is used for dimensionality reduction.The experimental results show that the fusion features can effectively improve the recognition rate;The weighted voting method is used to fuse lpdtp,mcs-ldp and cs-ldp at the classifier level.The experimental results show that the multi feature fusion through weighted voting can also improve the recognition rate.(4)The facial expression recognition system is designed and developed.QT is used to complete the development of the system and user interface.The input pictures are processed by calling the image processing algorithm of OpenCV computer vision software library.The system consists of three main modules: data acquisition,face detection and expression recognition;realize the functions of reading images,videos and cameras and facial expression recognition...