Facial Expression Recognition Based On Local Magnitude Coding

Facial expressions are the most direct carrier of emotion perception and the most obvious way of emotion expression.As intelligent robots,we want them to have a friendly appearance,and also want them to have human-like emotion perception and expression capabilities.Although many classical and effective facial expression recognition algorithms have appeared,the research on facial expression recognition has not yet entered the mature stage.In summary,research on facial expression recognition algorithms has important theoretical significance and practical value for improving the friendly interaction ability of intelligent robots.This dissertation mainly explores and experiments from two aspects of feature extraction and classification recognition.Improve the classic local texture feature extraction algorithm and classic local geometric feature extraction algorithm.Combine meta auxiliary learning network and SVM in the design of classification recognition.The specific work is as follows:(1)The chapter 3 presents a new descriptor,Center-Symmetric Local Signal Magnitude Pattern(CS-LSMP),which takes signal and magnitude information of local regions into account and greatly reduced feature dimensions compared to conventional LBP-based operators.Additionally,due to the limitation of single feature extraction method and in order to make full advantages of different features,our employs CS-LSMP operator to extract features from Orientational Magnitude Feature Maps(OMFMs),Positive-and-Negative Magnitude Feature Maps(PNMFMs),Gabor Feature Maps(GFMs)for obtaining fused features.Unlike HOG,our work generates Orientational Magnitude Feature Maps(OMFMs)by expanding multi-orientations.Our build two distinct feature maps by dividing local magnitudes into two groups,i.e.,positive and negative magnitude feature maps.The generated Gabor Feature Maps(GFMs)are also grouped to reduce the computational complexity.The experimental results show that the algorithm and framework proposed in chapter 3 have achieved significant classification results compared with some advanced methods.(2)The chapter 4 analyzes the advantages and disadvantages of the OMFMs proposed in the third chapter,and then proposes the Improved Orientational Magnitude Feature Maps(IOMFMs).Compared with the original OMFMs,the IOMFMs further divides the local orientation into a primary direction and a secondary direction and the primary direction is no longer limited to local edge pixels in the calculation of directional amplitude,which makes the extracted local information more complete.At the same time,the local gradient direction is added in the pattern,and the main pixels in the neighborhood are also considered in the calculation of gradient magnitude.In addition,in the face of the shortcomings of a single SVM classifier in classification recognition,the chapter 4 firstly introduces the Meta Au Xiliary Learning(MAXL)network to perform feature extraction and rough classification on facial expression images,and then use the improved feature extraction operator and SVM classifier for feature extraction and fine classification.The goal of MAXL is to automatically discover these auxiliary labels using only the labels for the primary task.In other words,the method automatically learns appropriate labels for an auxiliary task,such that any supervised learning task can be improved without requiring access to any further data.The experimental results prove the effectiveness and superiority of the algorithm and recognition framework proposed in chapter 4.