A Human Face Detection For Complex Backgrounds

Many people have done much work on Face recognition. But in this field,we have not overcome this problem until now. It reflects the limitation of our knowledge to ourselves. In these days, there are much development in computer technology and may new algorithms is emerging. We have the hope to overcome this problem again. This dissertation mainly concentrate on face detection.Face detection is the previous step of face recognition in this dissertation. We propose two ways to detect face. One is PCA based on skin color, the other is Log-Gabor based on skin color. These two ways can detect face on time with satisfied accurate rate.We utilize three color spaces to label pixels as skin including RGB,YCbCr and H.SV. The last two spaces give the better result in our experiment. So we have candidates to be verified after color segmentation in HSV space.Skin color alone is usually not sufficient to detect face. Next, we verify these candidates by searching for facial feature such as eye. We utilize two ways to detect eye based on the previous result.First, we applied principal component analysis to face detection. PCA on a training set of eye images is performed to generate the Eigenpictures which span a subspace of the image space. Images of eye and noneye are projected onto the same subspace and clustered separately, then we can separate eye from noneye. In our exeperiment, we perform principal component analysis through SVD.Second, we applied wavelet representation to face detection. We propose Log-Gabor detection based on skin color. Gabor function is the only one that can get the lower bound of calculation uncertainty and it can achieve good result both in the spatial and frequency domains. Our experiment result also shows that Log-Gabor method base on PCA gives is more efficiently in facial feature location.