Face Recognition Based On Statistical Features

Because there is a great application foreground in many fields, such as security verification, video surveillance and home entertainment, face recognition have gotten a lot of concern from researchers in recent years. And, it has become a popular research topic in the field of biology feature identification. In fact, during identification, compared with other ways of biology feature identification, face recognition is the most convenient and direct means. In the paper, we mainly study face recognition technology, and focus on the way for face recognition by statistic features. The innovations of this paper are as follows:1, Two face recognition methods based on wavelet transform and multi-feature are proposed. The first one combines the multi-feature with the multi-classifier. The second is using multi-feature to construct hybrid feature, and then introduce SVM to classify these hybrid features. Both these two methods performed very well in our experiments. .2, A new face recognition approach based on facial local constituent feature fusion is presented. Firstly the face image is segmented according to the facial prior knowledge, and the eigenface feature of each constituent feature is extracted, then the fusion and classification are implemented by SVM and Boosting algorithm. The experiment results proved that the performance of this method was better than that of traditional methods, which use the feature of whole image directly.3, The spectrum face acquired by combining the wavelet and frequency spectrum has great merit in face recognition. A face recognition method based on spectrum face is proposed in this paper. The spectrum face of original image is computed via wavelet transform and Fourier transform, then the eigenface and LDA features of spectrum face are extracted as classification feature, and several classification methods are adopted to get the recognition result. The experiment result proved that the method based on spectrum face could effectively improve the recognition performance comparing to the method using original image in spatial field. This method was called Spectrum Feature Face method. Besides, a face recognition algorithm using multi-feature and Radial basis Function Network (RBFN) based on this feature extraction is proposed, and it was used to tackle the problem of single training sample.4, In order to overcome the problems caused by small-scale variations of pose, facial expression and illumination in human faces and improve the performance of face recognition systems, a two-stage face recognition algorithm using MutualInformation and hybrid feature is proposed, and the approach is tested on several face databases. The experiment results demonstrated that the method could keep good recognition performance even if there are some small-scale variations of pose, facial expression and illumination in human faces.5 > The variation of illumination in face image always brings difficulties to face recognition. To tackle this problem, two face recognition methods aim to different illumination variation conditions are presented. The first is illumination subspace method; the second is an approach to increase training samples by using RBFN to produce face images or features under virtual illumination, and we call it Virtual Face method. The methods can both solve the problems caused by illumination in face recognition.