Based On Wtpca Face Recognition And Third-order Neighbors

The face recognition is the overlapping domain of the pattern recognition and computer vision, which widely applied in the discipline of robots. Compared with other living characteristics, the face recognition has the characteristic of directly, friendly, and conveniently, and easily accepted by the customer, therefore, it becomes the hot spot research in the current pattern recognition and artificial intelligence domain.In view of distinguishing the present situation of face recognition, this topic proposes the face recognition algorithm based on the wavelet transformation, principal component analysis and three-neighborhood classification. First, using the wavelet transformation to withdraw the low frequency coefficient, then withdrawing the main characteristics of the face images from the low frequency coefficient using the method of principal component analysis, finally distinguishing the respective category of the recognition images using three-neighborhood classification. The algorithm improves the recognition rate of face images and does not affect the computation speed and computation quantity.After preprocessing face images we can effectively remove the useless and disturbance information of the primitive images, and the information is difficult to withdraw. We can improve the image quality. The preprocessing technology of face images which this article introduces mainly has the filter to go to noise, the gradation transformation, the edge detection, the normalization, the gradation interpolation and so on.After preprocessed the dimension of face images is big, and we need to decrease the dimension in order to reduce the computation quantity. After wavelet decomposition of the face images, the low frequency has concentrated most information of the primitive face images, and the high frequency has mainly manifested some details of the primitive face images. Therefore, this article uses the two-dimensional separate wavelet transformation which is called db2 wavelet to decrease the dimension of face images. The method both can reduce the dimension of face images easily, and can withdraw the characteristics of the face images easily. After wavelet transformating the dimension of face images is still big. This article further reduces the dimension of face images using principal component analysis method, which causes the transformated lower dimension space still have good face expression ability, and projecting the training face images to lower dimension space forms training face database. When recognizing the face images, each recognized face images are all needed to preprocessing, wavelet transformating, characteristic withdrawing, and projecting to lower dimension space, then compared with the training face database to obtain the respective category. This article uses three-neighborhood classification, which can distinguish the face images of different expression, also can distinguish the face images of different posture and different decorations. The method enhances the recognition rate of face images.Through massive simulation experiments, this article concludes that using principal component analysis for low frequency containing most information of primitive face images, using three-neighborhood to classify face images for the withdrawn characteristics, the method obtains much better recognition rate, and decreases the computation at the same time, compared with other methods. The experiment indicates that, in the ORL face database having face images of 40 people, the recognition rate may arrive 98%, when using the former 8 face images of each person to be trained, composing the training set having 320 face images, and using all face images as the testing set.