Human Face Static And Dynamic Detection

Face detection is originally derived from face recognition, which is the key section of automatic face recognition system. In recent years, face detection has been attached much importance to by researchers as an independent task with the development of electronic commerce.Nowadays, the application background of face detection has exceeded the range of face recognition system, and it has important applied value in Content-Based Image Retrieval, Digital Video Processing and Visual Supervision And Control etc.The research of face detection advanced from simple image process to complex real-time video process. However, face detection has various weaknesses, such as large computation, low speed, and high false-detection rate, due to the complexity and frangibility of human face; but, Viola present a fast face detection method that is the most advanced method in face detection field based on AdaBoost learning algorithm in 2001. The method uses integral image to quickly calculate the feature, and construct weak classifier by the feature; then a strong classifiers is generated from diverse weak classifiers via AdaBoost self-learning algorithm; at last it uses the cascade structure to synthesize a more complex cascade classifier from single classifier, rapidly discarding the non-face area and guaranteeing the detection speed.In this paper, the main algorithm skeletons are the AdaBoost study training algorithm and Cascade algorithm sorter. This paper makes certain improvement to the slow training characteristic of the AdaBoost, uses a simple and Cascade algorithm classifier which filters the sub-windows impossibly including human face from an abundant of detected sum-windows, and retains sub-windows more possibly including human face. These simple sorters are arranged in the early stage of the multiple constructions, enabling the sub-windows separated from the non-face area to be filtered after much less arithmetic steps, thus avoiding multiple layer complex arithmetic. But the later-period sorter is designed for reducing the error-judging rate, and the Cascade algorithm skeleton and multiple structure classifier is able to decrease the arithmetic time and enhance the detection speed at the same time. Celeron 2.8GHz PC, face detection proceeds 15 frames per second.