Study On Multi-Position Eye Location And Facial Expression Recognition

With the rapid development of computer technology and biomedical engineeringtechnology, to use of human biological characteristics to study biometric recognitiontechnology, such as human identity, voice, fingerprint, facial expression, were becamethe preferred method in secure authentication. In the field of artificial intelligenceresearch, as an important biological characteristics, face detection, face identificationand expression analysis play a key role in the realization of machine intelligence, has abroad applications.With the tremendous progress in image processing and pattern recognition,processing technology on human face has been well developed, but there still existreality and uncertain factors, such as light, gesture, expression, occlusion and so on.Research on facial expression recognition was impacted by different people in differentdegrees with same expression and different expressions from same person, theories andmethods need to be perfected. Complete facial expression recognition including facedetection and positioning, feature extraction, dimension reduction and expressionclassification. The main contents of this thesis includeing multi-position eye locationand facial expression recognition, as follows:1. To solve the problem of eye location of human face in multi-position andcomplex background of color image, a method for eye location based on skinsegmentation and Gabor filter is proposed. Firstly, the improved otsu algorithm basedon particle swarm optimization is used to segment the image and candidate regions offace are obtained; Then an Gabor filter is devised to filter the image to highlight the eyeregion; Finally, the gray projection was used to determine the position of eyesaccurately. The results show that the method of skin segmentation is work well and hasadvantages about eye location of human face in multi-position and complex background,which is important for face recognition in real time.2. PCA face recognition algorithm was studied and apply it to facial expressionrecognition. The method assumes that the input data to be global linear structure, sowhen the data on the nonlinear manifold, the processing results will be affected. Inaddition, nonlinear feature extraction methods KernelPCA, LLE and KernelSLLE werestudied, and these four methods were implemented on JAFFE face database. The resultsshow that nonlinear dimensionality reduction algorithms have better results in facial expression recognition, it show that face space may be a high-dimensional nonlinearsubspace, which is located on a nonlinear manifold.3. A algorithm integration of local features for facial expression recognition isproposed. The greatest contribution to facial expression recognition are human eyes andhuman mouth, so integrating these two aspects to facial expression recognition, at thesame time given these two aspects different weight coefficients. Compared with theoverall characteristics feature extraction,not only reduce the computational complexity,but also reduced the expression discrimination error caused by the similarcharacteristics of same person and different people in different degrees, improve therecognition rate.