Research On Fast Face Detection Algorithm Based On Skin Color Information

Automatic recognition of human faces is an important research area of computer vision and image understanding in recent years. Human face detection that acts as one part of the automatic human face recognition system and is responsible of locating the faces is the prerequisite for the whole system to ensure it works normally and high efficiently. Due to human face' s variability, human face detection becomes a rather complex issue of pattern recognition. Recently, because of the potential value in most areas, such as security surveillance, content-based image retrieval, etc, human face detection has become a single research topic and attracted broad attention of more and more researchers.As for face-detection methods, such as appearance-based method, feature-based method, template-based matching method, skin color-based method etc were presented before. Among these methods, thejnethod of using skin color information to detect human faces is very saitable for locating faces in automatic face recognition system with the advantage on directness, rapidity and simplicity.In this paper, skin segmentation algorithm is presented based on statistic color model of human face and different color spaces. The color spaces of YCbCr, HSI, KL, YUV, and YIQ are involved in this work. The detail is to transform the original color image from RGB space into abovespaces, and then select a great amount of skin color pixels manually to establish skin color model and find the scope of skin color.Based on the research with skin color model, novel algorithms are further proposed by me to implement fast and real-time face detection based on skin color information with two kinds of different application scenarios. One is to detect single face in a simple background image and the other is to detect multiple faces in a complex background image. The proposed algorithms handle differently according to the different application feature. As for the single face detection with simple background, the method of projection is mainly used in the algorithm. First, segmenting the color image into skin color and non-skin color regions, then implementing projection in horizontal and vertical direction respectively within the skin color regions, the face could be located by the result of projection. But as for the multiple faces detection with complex background, region labeling algorithm and face conformation based on knowledge rules are mainly used instead. First, skin color model is still used to separate the color image into skin and non-skin regions, classifying the different skin regions with region labeling algorithm, deciding the candidate human faces based on the proposed knowledge rules within all skin regions, thereafter, transforming the color images within the candidate face regions into gray images and comparing them with the proposed rules of facial gray level distribution, implementing face verification and Locating faces according to the result. The algorithm analyzes combining the color with gray space, eliminating the things that have the similar color with skin color but big difference in gray level distribution. Many advantages of real-time, fast detection and without the restrain of size, posture and expression are still possessed in the algorithm. Because the majority of existing face detection methods can' t realize that, it would be have high application value and meaningful to the future related research work.Experimental results reveal the validity of the proposed approaches by me in solving face detection problem and the approaches are appropriate to the step of locating faces in automatic face recognition system.