Expression Recognition Based On Deep Learning

Facial expressions are able to express human's emotions and intentions,so facial expression recognition,which is the important foundation to achieve the human-computer interaction.Traditional facial expression recognition relies on a complicated feature extraction algorithm designed carefully by people,but also partly loses the original facial expression feature information.In recent years,as a feature learning algorithm driven by pure data,the deep learning can automatically learn more essential characteristics of sample.So,the method and application of facial expression recognition based on the deep learning is discussed and studied in this paper.The main research work and results are summarized as follows:(1)The facial expression database is augmented.In this paper,preprocesses such as face detection and normalization are conducted to the existing facial expression database,and data augmentation strategy is applied to augment the facial expression database.(2)A facial expression recognition method based on the Deep Belief Networks(DBN)is studied.DBN model parameters are adjusted and optimized through process of pre-training and fine-tuning,and finally the classification results of expressions are outputted.The recognition rate of 91.16% is achieved on the CK+ database.(3)A facial expression recognition method based on Convolutional Neural Network(CNN)is studied.Changes in facial expression are often subtle,while CNN could capture the local feature of images,which is fit for classification and recognition task of two-dimensional face images.By this method,the recognition rate is increased by 5.02%.(4)A facial expression recognition method based on NIN(Network in Network)is studied.The convolution layer in NIN have stronger nonlinear feature extraction ability,thus being advantageous to the abstraction and expression of nonlinear features of complex facial expression images.This method further improves the recognition rate by 2.79%.(5)A facial expression recognition demo system is implemented.The system is mainly divided into two functions.To recognize facial expressions automatically,dividing them into seven discrete emotions.To visualize convolution layer,observing the feature map generated by each convolution layer after convolution operation.