Research On Algorithm Of Facial Expression Recognition Based On CECBP And The Decision Of Salient Region Features

Facial expression recognition is a very active research topic in the frontier fields of computer vision, artificial intelligence and pattern recognition, and it also has very broad application prospects in intelligence human-computer interaction and intelligence home furnishing fields. The technology of facial expression recognition also has many important practical applications in real life, such as medical and information security. Because the field has the very important theory significance and market value, more and more scholars have joined in the research on the technology of facial expression recognition.Facial expression recognition system can be divided into three stages: face detection and image preprocessing, feature extraction and classification. Along with the research of facial expression recognition unceasingly thorough, some problems in the feature extraction appear constantly, such as the robustness, real-time and stability in the feature extraction algorithm. This dissertation studies the traditional feature extraction algorithm, which is based on geometric features and texture features, focus on the analysis of the advantages and disadvantages in the feature extraction algorithm based on texture features, and aim at the limitation of traditional algorithm to put forward the improved algorithm. The main work and contributions are as follows:(1)Through the study of the feature extraction algorithm based on texture features, and summarized the insufficiency of the existing feature extraction algorithm, a feature extraction algorithm based on multi-scale Center Error Compensation Binary Pattern (CECBP) is proposed. In the algorithm, the images are firstly preprocessed and multi-scale pyramid of these images are then created. And then, CECBP is utilized to encode the images of every layer in image pyramid. Finally, the chunked and encoded histogram sequences are used as a feature and Support Vector Machine (SVM) is used in classification. To demonstrate the superiority of the proposed method over traditional Gabor wavelet and LBP, cross-validations on JAFFE, Cohn-Kanade and Pain expression database show that the method can not only suppress noise, but also have state-of- the-art recognition accuracy and faster recognition speed.(2)After a deep research on the technology of facial expression recognition, in order to further enhance the recognition accuracy and improve the robustness of facial expression recognition system, image salient region and HOG features are introduced on the basis of CECBP and the decision of salient region features is proposed. Firstly, according to the information richness in different area of the face, the image salient regions are located based on the entropy of information. And then the CECBP features were extracted in these areas, the HOG features were extracted in the whole salient region. Finally, these features are input into SVM, according to the different weight of each classifier and decision-making mechanism, the final recognition result is determined. The cross-validations on JAFFE, Cohn-Kanade and Pain expression database show that the method make full of the rich expression information of the salient region, not only inherits the superiority of CECBP, such as suppress noise and effectively extract texture information, but also resolve the limitation of the single features, further improve the recognition accuracy and effectiveness.