Research On Facial Expressions Recognition Method

Facial expression recognition is a new research direction, and it has broad applicationprospects in the intelligent human-computer interaction. At the same time, facial expressionrecognition is widely applied in the fields of the traffic, medical treatment and public safety.In recent years, facial expression recognition has been paid close attention by more and morescholars. And this technology has become a hot research in the area of artificial intelligence.Therefore, research on facial expression recognition technology has important theoreticalsignificance and practical application value.Facial expression recognition technology includes image preprocessing, featureextraction and classification. In image preprocessing, a new eye location method based onSobel edge extraction was proposed. The new method includes three steps for coarsepositioning of eyes, segmentation of eyebrows and eyes, and precise position of eyes. Aftermedian filtering and normalization for image, the coarse area of eye was located according tothe priori knowledge. The segmentation of eyebrows and eyes was implemented according tothe segmentation line of eyebrows and eyes, and the line was determined by the gray integralprojection curve. Then, the precise location of the eyes was realized according to theboundary of binary image obtained by extracting eye edge with Sobel operator. Comparedwith the traditional template matching and circle examination of Hough transform, the methodproposed has obvious advantages in computing time and is suitable for real-time system.In feature extraction, the feature extraction method based on the Gabor wavelettransform and Non-negative Matrix Factorization was proposed. Considering thecharacteristics of Gabor wavelet transform，corresponding Gabor filters were designed. Thefacial information area was filtered by Gabor filters, and different sub-picture information wasobtained. Then, sub-pictures generated from each filter were decomposed by non-negativematrix factorization, implementing data dimensionality reduction and feature selectionprocess.In classification, the two-layer classification model based on nearest neighborhoodclassifier and the probability statistics was designed. As input information, sub-pictureinformation achieved from each filter was input the nearest neighborhood classifier, whichwas regard as the first layer classifier. The output results from the first layer classificationwere calculated probability to implement the second layer classifier. The output results of thesecond layer were the final results. Under the design of the two-layer classification model,each image was distinguished two times, reducing the possibility of misclassification andimproving the robustness of the algorithm.Under the MATLAB programming environment, the JAFFE facial expressions databasewas applied in experiments. Experimental results showed this method is effective. In addition,the method had also been applied in face recognition with small samples. Related experiments were done in the Yale face database, and the results showed the method proposed is alsosuitable for face recognition.