Cross-domain expression recognition based on representation self-challenging
Domain difference is one of the major obstacles when transferring facial expression recognition systems from one data set to another.However,current algorithms are usually trained and evaluated on a single data set,thus they cannot identify expressions in different styles and environments well.In order to solve this problem,the domain generalization algorithm is applied to the facial expression recognition task.Experiments are conducted using a variety of public facial expression datasets.Compared with the general deep learning and even transfer learning methods,the ResNet-50 is used as the backbone network and unify the parameter settings,the corresponding test set is not adjusted,and the representation self-challenge method is adopted for training.Experiment results show that the average recognition rate of the method in the test data set reaches 63.46％,which is better than the general domain adaptation algorithm.The evi-dence shows that the method helps to extract more general features in expressions,thereby improving the classification accuracy.
万方数据
维普
