Research Of Face Detection Based On Improved AdaBoost Algorithm

Face detection is the first step in face recognition, which plays animportant role in embedded control systems and other safety systems. For any imagewhich is given, it is easy for people to find the face’s location by using eyes. Butcomputer must do some calculations so can it locate the human faces. If a humanface exists in the image, calculating its position and size. This process is called facedetection.Because AdaBoost algorithm has some advantages such as high detection rateand fast detection speed and so on, it becomes one of the most successful areas of theface detection algorithm. However in practical applications, it is important toincrease the rate detection of AdaBoost algorithm. This paper focuses on the facedetection phase. By studying which factors influence the face detection rate and falsedetection rate, proposing a method to quickly and accurately detect human faces inimages.(1)By studying the process of AdaBoost learning algorithm, it finds some factswhich influence the detection rate. Download pictures from the internet randomly.The images are all640x480pixels. According to these images, the face size ofGaussian model and the single face weight model are established. The establishmentof these two models can help to analyze human face distribution trends. Using themin the process of face detection can help to reduce duplication of testing anddetection of invalid.(2) According to the face size of Gaussian model and the single face weightmodel, a new scaling mode is established when doing face detecting. This methodcan quickly determine the most likely range of face sizes appear. According to thisrange, it can help to set the scaling factor. Experimental results show that this methodcan effectively improve the detection speed for the single face image.(3) By analysis the characteristics of faces, designing a model for BP neuralnetwork detection algorithm which based on the optimizing AdaBoost. In this mode,a hierarchical way in the BP neural network is used. The hierarchical way is set by the face size of Gaussian model. First, using the optimized AdaBoost algorithm todetect the original image and getting the range of face region. Then using the BPneural network classification to determine the face’s range and get the final result.