Study Of The Methods For Human Face Detection And Its Application To Video

Today in information processing fields of human face images, there are manyresearch directions such as face recognition, face tracking, pose estimation,expression recognition and so on, face detection and location are involved in all ofthose applications. Along with the generalization of computer application,theimprovement of computer performance and achievements in image processing andpattern recognition fields,more and more emphases are put on the research onhuman face detection. In this thesis, face detection under complex background ismainly investigated.On the basis of the analysis of existing techniques, a Gaussian model and neuralnetwork based human face detection method is presented for color images undercomplex background. First, Gaussian models are applied to fast extracting ofskin-color and lip-color regions. Then, face candidates obtained are rotated in imageplane to be upright. Finally, a BP neural network is used to determine whether acandidate area is a human face or not. Simulation results show that the proposedapproach is of high precision and generality. As widely used in video processingfield, face detection technology is introduced into the field of anchorperson shotdetection for news video indexing. First, a representative key frame is extracted fromeach news video shot. Based on the similarity among the anchorperson key frameswith the same model, the graph-theoretical clustering algorithm is applied to groupthe key frames into anchorperson clusters labeled with different anchorpersonmodels. Then, a classifier based on fuzzy if-then rules is used to further tune thedetection results. Experimental results demonstrate the efficiency and robustness ofthe detection scheme.The detection methods proposed in this paper have excellent performances andhigh practicality.