Expression Recognition Based On Twice Feature Extraction

The research and application of facial expression recognition algorithms are hot topics studied by scholars.With the maturity of technology,the application range of expression recognition is becoming more and more extensive.In the past period of time,the expression recognition algorithm has reached practical value in some aspects such as recognition speed and accuracy,however,due to the harsh laboratory environment and low diversity of the samples,the algorithm in practical application shows performance in terms of robustness and generalization,and there is still much room for improvement.It is of great significance to improve the robustness and generalization of facial expression recognition algorithms for improving human computer interaction methods,analyzing human emotions and medical research applications.In order to solve the problem that the existing facial expression recognition algorithms have poor recognition effects on individuals outside the standard expression database,this paper improves the expression feature extraction and facial expression recognition algorithms respectively.The algorithm includes four parts: expression database construction and preprocessing,expression feature extraction,expression feature secondary extraction and classification,and network weight configuration.Firstly,the database is constructed by combining standard expression database,self-collected expressions,and expressions collected on the Internet.Secondly,the LBP features are improved by combining the global image features and local features to provide a more robust adaptive threshold LBP feature.The threshold LBP feature and the Kirsch-Canny feature are used together as the initial feature.Then,two three-layer DBN networks are used to extract the features of the initial features and the primary classification is performed.Finally,the verification set data is used to determine the weights of the two DBN networks.The weighted result is the expression recognition result.Experimental results show that the recognition rate of the proposed algorithm in the CK+standard expression database reaches 95.2%,and the recognition rate of the same model in theself-made expression database reaches 81%.The comparison experiment with other algorithms shows that the recognition rate of the experimental individual in the CK+expression database is 0.6% lower than that of other algorithms,but the recognition rate of the expression individual outside the standard expression database is increased by 14%.