The Location Of Object Feature And Geometry Adjustment For Face Recognition

Now, the face detection and recognition has already became the important research topic in the machine vision domain. It is applied in the harmonious man-machine interface, the vision detection, digital video frequency processing and etc. Until now, many algorithms of the face detection and the recognition has already proposed ,and with which, the good effect was obtained. But the velocity of human face detection is need to be researched cannily as well as the middle part of face detection and recognition which is about the selection of face features and the face alignment. They are precondition for setting up the fast system of face recognition.Compared to paying great attention to the rate of accuracy before, the research of this thesis cares velocity more. Taking the low distinguishing rate image as object, first,we detect face using the cascade classifier based on Adaboost statistical learning method to detect face fast. Then,in order to make the face image founded standard, we propose a fast feature detection method based on edge detection and gray grads density. The method can locate eyes which is the main feature of face,and based on which, we can locate mouth using geometry and grads information. At last,according to the linear relations between shape and texture which is proposed by DAM(direct appearance model),we train the transformation matrix by doing statistical study with a lot of sample labeled to locate the face direction and geometry adjusts.The feature detection method proposed in this thesis is based on edge detection and gray grads density. It can locate eyes quickly and exactly in some sense, and adapt to some complex influences(non-glare condition).Face images labeled could provide the training sample data with face alignment, and also could apply to the location of face direction and geometry adjusts preparing for the face recognition well.