Extraction And Analysis Of Facial Expression Based On Image Recognition

With the updating of science and technology products,how to become more intelligent for human-computer interaction becomes a very important research topic.Emotion is the most intuitive performance of human communication,so in the human-computer interaction,the emotion becomes a kind of very important information resources.Usually the study of emotional computing includes three aspects: facial expression recognition,speech recognition and human behavior recognition.In this paper,the main content of the research is the expression recognition.In addition,with the popularity of multimedia devices,the growing number of online video,through the characters' emotion to classify video can strengthen the management of network video and enhance the audience's user experience.The main contents of this paper are as follows:1.Image normalization.The extraction of facial features is first required to locate the face area.However,there are a lot of image noise come from background,light and image size in the collected sample data.And face area may not alignment because of image rotation,so we need to preprocess image samples and get face area before the expression recognition.This paper use Adaboost algorithm to detect the face from the image,and then do correction for tilt face image based on binocular coordinate position before the extraction of facial features.2.Face features extraction.the geometric features and texture features from image samples are used for expression recognition.Texture features are extracted with Gabor wavelets and uniform local binary model operator(ULBP).Geometric features are extracted by using active shape model(ASM).This paper summarizes the key feature points for expression recognition by computing the distance between the geometrical feature points of the face.3.Face features dimension reduction.Gabor wavelet texture features extraction will lead to high features' dimensions and memory space consumption,so this paper use the sparse method and the principal component analysis(PCA)to reduce features' dimensions.the accuracy of this method were compared and analyzed on the data set.4.Facial expression recognition.Facial expression recognition is a multi-classificationproblem.In this paper,we use the optimized support vector machine(RB-OVR-SVM)algorithm based on binary tree to realize the recognition of facial expression.5.Video classification.According to the research of facial expression recognition method,the face detection and facial expression recognition are carried out on the video clips.According to the existing psychological and probabilistic information,we can sum up the emotional characteristics of various video segments and classify the video.This paper puts forward the following innovations and improvements during the research process:(1)The combination of texture feature extraction and geometric feature extraction is used to extract facial features.(2)In order to improve the computation speed and ensure recognition accuracy,the sparse feature selection is used to reduce the dimensionality of facial feature.(3)The support vector machine(SVM)algorithm based on binary tree is optimized and used for facial expression recognition successfully.Compared with the previous support vector machine for multi-classification algorithm,it not only improves the speed of expression recognition,but also solves the blind problem of data indivisibility.In this paper,the texture feature and the selected geometric features are extracted simultaneously,and the dynamic information of the image is fully taken into account.At the same time,using the RB-OVR-SVM algorithm,the classification accuracy is 94.4%.