The Method Of Face Detection And Recognition Based On Multi-features Fusion

Face detection of color images with complex background and face recognition of gray images were studied in this paper. A new algorithm for human face detection based on skin color and feature verification was presented. The algorithm was composed by three parts, skin color segmentation, excluding false area, and feature verification. Firstly, how to select the color and how to make the model of skin color in the color images with complex background were discussed in detail, an adaptive threshold values algorithm was proposed. Then excluding some false areas based on knowledge verification, we can verify face based on eye and mouth by using edge detection, template matching and projection. Finally, nose can be detected by projection approach. We experimentally evaluated the performance of the algorithm using a variety of human face. Our results show that the algorithm has high accuracy, robustness, and low computational cost. After detecting faces, we normalized and grayed color faces, an algorithm which combining Gabor wavelet transform with the direct linear discriminant analysis was presented. The algorithm gained higher correct recognition rate using less number of feature vectors by getting the feature vector module as Gabor feature by Gabor transformation in areas, extracting feature by linear discriminant to reduce the space of feature and to make the classifier more compact and effective. As the experimental results demonstrated the algorithm not only gets high accuracy and robustness to the varying of illumination and expression, but also performs better than traditional Fisher Linear Discriminant Analysis and Gabor wavelet transform plus LDA approaches.