Study Of Automatic Face Detection In Complex Background

As a critical technology of face information processing, face detection has been paid much attention and become a very active research branch in pattern recognition and computer vision application areas. With the development of intelligent information processing, face detection will be broadly applied in identity recognition, content-based retrieval, surveillance and human computer interaction.The main work of this paper is to study knowledge-based method and statistics-based method on face detection. Knowledge-based method is to extract facial feature and set up some rules, then judge whether the image contain faces or not. Statistical-based method relies on numbers of typical samples to get statistical feature and set up a classifier which can classify face samples and non-face samples correctly.As to knowledge model-based method, an improved face detection algorithm based on feature extract is presented in this paper. Adaptive entropy threshold has been adopted to extract facial hengwen feature, candidate faces are obtained according to their spatial relationship, and they are confirmed by using gray and texture detection. Experiment results proved that hengwen feature will be more accurate by using dynamic threshold, which simplifies the calculation complexity.As to statistics-based method, a face detection algorithm based on multi-method optimal fusion is presented in this paper. It is a hierarchical approach and integrates skin color model, neural network and template matching effectively. In color image, skin color model is used for segmenting regions in which may have faces, and then two BP neural networks in cascade connection serve as a face detector, at last template matching is used to exclude false alarms. With the elimination of false areas, the research area will become smaller and smaller, one or multiple faces detection will be accomplished eventually. To detect faces with different sizes, a strategy of multi-resolution sliding window is used to accomplish invariance of location and scale in face detection. Experiment results show that the algorithm can detect frontal faces with complex background; it is a robust and effective algorithm.