Facial Expression Recognition And Application Based On Deep Learning

Facial expression which is one of the most important ways of human emotion,is a hot research field in natural human-computer interaction.It is widely used in video conferencing,interactive games and virtual reality.At present,the feature extraction of facial expression recognition(FER)mainly adopts the feature of artificial design,and the features are independent and unrelated.It is difficult to reveal the essential features of expression comprehensively and effectively.Deep learning which is different from the traditional way of facial expression recognition based on facial appearance or geometric features has the ability to learn features from the data independently.In this paper,deep convolutional neural network in deep learning is used to study facial expression recognition and applied to the natural interaction system of virtual environment.In this paper,we mainly study the expression recognition algorithm based on single-channel network and parallel network.we augment the data in the Data Set by setting the region of interest(ROI),and consider that the softmax classifier is not ideal,so we use SVM as the classifier.We combine the data without preprocessed and use the maximum confidence way to combine the single path network to form a parallel network in order to further enhance the recognition performance of the model.The main contents are as follows.1.Expressions recognition based on single path depth network,Under the framework of Caffe deep learning open source framework,the construction and implementation of the network structure were completed.In order to seek a single-network structure with better recognition performance,three comparison experiments were designed respectively.The contrast experiments with Different data enhancement methods,different classifiers and different pretreatment methods propose a deep learning expression recognition method which is the Phase combination of LBP+DCNN+SVM.By comparing with the traditional method of machine learning and other deep learning modes,the robustness and accuracy of the proposed algorithm are verified.2.Expressions recognition based on parallel depth network,Although the single channel recognition network has achieved a good classification effect,the characteristics which learned are relatively single.In order to enhance the performance of network identification model,the performance of parallel networks under combinations of different input features is further explored,and put forward the parallel network identification method of the combination of the two input features--LBP and Depth.Compared to single path networks,it helps to improve the classification ability of the recognition model.3.A brief study is conducted on Virtual environment natural interaction system scene and character modeling.Through geometric,physical and behavioral modeling,a three-dimensional virtual learning environment is constructed to simulate a realistic and intuitive learning environment.4.Using deep learning network training to obtain the final expression recognition mode.Combining the somatosensory interaction technology with the virtual reality technology to simulate the scene of classroom and completes the design of virtual environment natural interaction system based on expression recognition.The interaction between the expression and the virtual human in the virtual environment reflects the natural human-computer interaction process.