Several Face Detection Method

Images containing faces are essential to intelligent vision-based human computer interaction, and research efforts in face processing include face recognition, face tracking, pose estimation, and expression recognition. However, all of these researching directionsinvolve in one problem-face detection and location, in other words, before this faceprocessing, we must know faces' locations and scales. Consequently, to build an automated face processing system which analyzes the information contained in face images, robust and efficient face detection algorithms are required.The research on face detection has lasted for more than twenty years. But, up to now, due to the complexity of the purpose such as the diversity of face patterns, variable lighting condition and so on, many researches can not resolve the problem completely even if they have studied it for long time. In this thesis, the author has done some work on the face detection.The work includes:(1) Face detection based on multiple templatesIn this thesis, the author has studied the performance of the method of face detection on multiple templates. In allusion to the drawbacks of the method which are the lower computing rapidity, and the sensitivity to the lighting condition, we present that we make the KL-based skin color coordinate system as the first step efface detecting. By this way, we can eliminate a great deal of non-face areas, and then use templates to search for faces in images, in the end, determinate whether the candidate area has a face or not according to validating rules. The result of experiments shows that the pretreatment using color information improves the performance of the method.(2) A face detector based on mixtures of linear subspacesOn the pattern recognition, it is very natural to view face detection as a problem of classification. Since the images of a human face lie in a complex subset of the image space that is unlikely to be modeled by a single linear subspace, we use a mixture of linear subspaces to model the distribution efface and non-face patterns. In the other words, we used Fisher Linear Discriminator to project samples from a height dimensional image space to a lower dimensional feature space. The results of experiments show this method of face detection is efficient regardless of their poses, facial expressions and lighting conditions. Furthermore, the method has fewer false detects than other methods.(3) A face detector based on a sparse networkIn 2000, Yang presented a SNoW-based face detector, this method get the best result of face detection in the world by common consent, which can be proved in some documents. For the training parameters α,β are vital to the performance of face detector, we present an amended Winnow learning algorithm. The experimental result tells us it is effective.