Facial Expression And Gender Recognition Based On Static Facial Image

Face is one of the most important human biological features, many important biological information is provided by it, such as facial expression, age, gender, identity, race. It gradually becomes a research focus of recent years that computer vision and pattern recognition researches are based on face images. As important parts of face recognition technique, facial expression and gender recognition techniques had been widely researched and studied. By applying the techniques, face recognition correct rate was improved, and study field was extended. So it is important and meaningful to do such research on facial expression and gender.There are three main parts of facial expression and gender recognition system, they are face image pre-processing, facial feature extraction, classifier recognition. Those are major research contents of the paper.Facial gender recognition correct rates are been influenced by face image pre-processing. The pre-processing includes face detection, scale normalized, gray-scale normalized. In the paper, face images are been detected by well-known cascade based on AdaBoost algorithm. Because the method's detection has higher speed and accuracy. Then the images are been processed by using scale normalized, gray-scale normalized methods.Facial gender recognitions include face feature extraction and classification, classification is much more important. It's the key to recognition that could extract interesting face features from high dimensional face images. Feature extraction strategy based on principal component analysis (PCA) was introduced, then feature extraction strategy based linear discriminant analysis (LDA) was introduced latter in the paper. PCA and LDA algorithm were combined, so it was got over that the defect of LDA algorithm that LDA algorithm couldn't be applied on small sample size problem, and face recognition correct rate was improved by PCA and LDA algorithm. Last, Euclidean distance was applied as similarity measurements of pattern features, classifier based on distance function was applied to classify expression images. The result of mean facial expression algorithm's generalizations were not desirable, that needs to be improved in future research. Gabor wavelet and elastic templates matching methods were also applied in facial expression recognition test. In the process of facial feature extraction aspect, we made a one by one pixel two-dimensional wavelet transformation to the image firstly, and then expression feature vectors of the expression sub-regions were extracted by wavelet transformation, at last we accounted the degree of similarity by Euclidean distanc. In the research of model-matching algorithm, we proposed a simulation model-matching algorithm. In the end we used the K-nearest neighbor classifier to achieve the seven kinds of basic facial expression recognition, the experimental results were better. However, the time complexity calculated by the K-nearest neighbor classifier was higher in the experimental process, it should be further improved.