The Algorithm Of Multiple Face Detection And Recognition In Complex Background

The face recognition technology becomes the most easily acceptable biometric recognition method because of its non-contact and concealment. So it gets comprehensive attention and a lot of research has been done in this field. But the results of detection and recognition are always affected by illumination, expression and occlusion et al. So it still has many technical problems to be solved.The thesis studies the algorithm of face detection and recognition with the static image in complex background. The face detection algorithm based on Adaboost and the face recognition algorithm based on LBP both have limitations. And the Gabor feature has high dimension and is not robust enough to the factors of expression et al. The thesis studies the solutions of these problems. Details are as follows:We combine the characteristics of skin color and grayscale with the traditional Adaboost algorithm. And the second face detection has been achieved by the skin color model and gray projection. The experimental result shows that the improved Adaboost algorithm can raise the accuracy of face detection.We use PCA to reduce the dimensionality of LBP operator to solve the problem that the dimension of LBP operator is too large when we extract feature. And the experimental result shows that the new LBP operator not only reduces the dimension and also make the new features more effective.The multi-scale and multi-directional Gabor filter can express the face features effectively, but it results in time-consuming because the dimension of feature is too large. We encode the eight Gabor Magnitude Pictures of each scale’s eight directions as one picture. The method greatly reduces the length of feature and improves the speed of recognition.The thesis proposes the method that we compute the LGBP histograms after partition the face image to reduce the impact of the face recognition by expression and occlusion et al. The experimental result shows the block model of 88 ? can guarantee the calculation speed and the same time obtain the high recognition rate.