Face Detection Technology, Based On Skin Color And The Adaboost Algorithm

With the continuous development of society and increasing of information's levels, people have higher requirements for authentication. Face is the important biological characteristics of the human body, which has great development potential in the identification, automatic monitor, human-computer interaction and many other areas. However, the face detection, which is the first step in face recognition, has become the more important topics in the field of pattern recognition and computer vision.Face detection means that certain method is used to search in any given image to determine whether any human face exists in the image. If it exists, the detected face location, size and pose are returned. The images originating from the natural environment are vulnerable to be affected by images'background, light environment and different poses of human. Thus the face detection becomes a complex and challenging research content. But face detection has wide application to security access, video processing and other areas. So it is favored by a large number of researchers.This thesis elaborates the basic concepts of face detection technology, and analyzes and discusses the realizing ideas of several common methods of face detection. Meanwhile, the thesis in-depth researches on the face detection technology. On this basis, the thesis proposes a face detection method based on the combination of skin color and the improved AdaBoost algorithm. Experiments show that the face detection method presented by us is reasonable and has theoretical and practical significance to some extent. This thesis's research work mainly focus on the following aspects:First, for the skin color information in color images, we give a method of skin regions detection based on the color space of YCgCr. This method uses Gaussian skin color model in YCgCr space to process the images to be detected, and get their skin color similarity images. And then it utilizes the adaptive threshold method to segment the skin area. At last, according to the facial features of face, mathematical morphology and the knowledge of connected component are adopted to carry on further treatment on the skin area to remove non-human facial region. Because the face detection method based on skin color is not susceptible to the impact of posture and facial expressions, so the method has higher practical value.Secondly, this thesis describes the traditional AdaBoost's learning algorithm. Focusing on the problem of degradation and distortion of sample's weights in training the weak classifier, this thesis gives some improvements on the rules of updating the weights. Experiments show that the improved method can prevent from the phenomenon of degradation to some extent, and improve the classifier's performance.Finally, for face detection method based on skin color has high false detecting rate under complex background, and AdaBoost algorithm takes a long time to achieve at face detection, a novel face detection method combining the information of skin color in YCgCr with the improved AdaBoost algorithm is proposed in this thesis. This method uses skin color information to segment the possible skin region, and then uses the improved AdaBoost algorithm to train to obtain a strong classifier to further verify the candidate face region. At last, the location of the human face is output and marked. Experiments show that this method can be effectively applied to face detection under complex background, and it has the advantages of high detection rate, low false detection rate and fast detection speed.