Research On Algorithm Of Facial Expression Recognition Based On Texture And Geometric Features

Facial expression recognition is an important part of affective computing and advanced intelligent, and it is also a hot issue in the fields of human-computer interaction, machine learning, intelligent control and image processing. The in-depth study of facial expression recognition has become much more important in order to promote a more natural and more humanized human-computer interaction. In recent years, there are still some problems of facial expression recognition, especially how to extract the expression features which have higher real-time performance, stronger robustness, more stability and more representative.The dissertation analyzed and researched on the texture features and geometrical features of facial expression, and proposed some improved algorithms in order to overcome the shortcomings of the traditional geometric feature extraction methods, the limitations of the traditional Gabor feature in facial expression recognition, as well as consider the advantages and disadvantages of different features in different environments. The main contributions are as follows:(1) Put forward a kind of facial expression recognition method based on the combined geometric features in order to overcome the shortcoming of the traditional geometric feature extraction methods. First, locate the key points of facial expression images by using the Active Appearance Model (AAM); Then, extract the Direct Geometric features (DG) and Indirect Geometric features (IDG), use Fisher Linear Discriminant (FLD) algorithm to reduce the dimension of DG features, and combine it with the IDG features, thus constituted the combined geometry features; Finally, classify the combined geometric features of facial expression images by using Support Vector Machine(SVM). The experimental results show that the combined geometry features have higher recognition rate compared with the traditional geometry features, and the features have a certain robustness to a little illumination change condition or gesture deflection condition.(2) Put forward a method of facial expression recognition method based on Gradient Gabor Histogram (GGH) texture features, aimed at overcoming the limitations of traditional Gabor features. First, extract Gabor features of the facial images after pretreatment; Second, construct Gabor features fusion images of the Gabor features which are extracted in the same scales but in different directions according to their gradient directions; Third, divide the fusion images into several blocks and calculate the distribution histogram of each sub-block, thus the GGH features of human face can be obtained; Finally, classify facial expression images by using SVM. The results of the experiments show that the real-time of the GGH features is better than the real-time of the traditional Gabor features.(3) Put forward a facial expression recognition method based on the texture features and the geometric features. First, locate the key points for facial expression recognition by using AAM, extract DG features and IDG features, and use FLD algorithm to reduce the dimension of DG features; Second, construct Gabor fusion images to extract GGH texture features; Third, classify the three groups of local features by using SVM to obtain three independent recognition results; Finally, the three independent results are fused and weighted in the level of decision to obtain the final classification result. The experimental results verify that the method can make full use of the texture features and geometric features of facial expression, and it can improve the recognition rate and the reliability.