Spontaneous Expression Recognition Using The Fusion Of Infrared And Visible Images

AS the rapid development of Computer Science and Technology and Artificial Intelligence, and the improvement of automation level, How to being more anthropopathic in communication with computer is paid more and more attention. The research of Facial Expression Recognition has such come out. It is focused by so many researchers because it's a subject crossed by Psychology, Engineering, Data Mining, Artificial Intelligence and so on. The research results of Facial Expression Recognition are quickly increased through the help of Face Recognition, Face Detection and Location. While many researches now are for the posed expressions, and few spontaneous expression recognition technology are researched. Also, the fusion of visible and infrared images may increase recognition rates as it contains more information that singal-light images. Therefore, this paper will do some research on spontaneous expression recognition using fusion images of visible and infrared images.In the paper, we proposed methods to recognize expression as below:1. we proposed a mathod of infrared facial expression recognition based on 2 layers 2D wavelet transform. It does 2 layers of 2D wavelet transform to images, chooses approximation coefficients as features by Largest Euclidean Distance, and recognizes the expressions by K-Nearest Neighbors. The algorithm is validated effect on both Equinox database and USTC-NVIE database.2. We proposed a method of facial expression recognition using fusion images of visible and infrared images based on standard co-training system. It extracts appearance features from visible images and approximation coefficients of 2 layer 2D wavelet from infrared images, and trains a standard co-training system to recognize expressions. It is validated that this algorithm can increase recognization rates of facial expression recognition under some conditions.3. We proposed a method of facial expression recognition using visible images and assistanted by infrared images. It extracts appearance features from visible images and approximation coefficients of 2 layer 2D wavelet from infrared images, first trains a K-nearest neighbor classification to recognize expression of visible images and then uses infrared images to adjust feature space of this classification. It is also validated effecter than facial expression recognition using single-light images only, on USTC-NVIE database.