Research And Improvement On Face Detection And Facial Feature Points Locating Technology

In recent years, face recognition, which is defined as a computer technology used in the identity Identification by analysing the visual information of facial features, has become a hot research topic in artificial intelligence. This technology can be widely used with Public Security Criminal Investigation detection, access control systems, video surveillance systems and network applications. Generally speaking, the face recognition system consists of image intake face locating, image pre-processing as well as face recognition. Face locating system consists of the face detection system and feature point locating system. This article will make an in-depth study in those two part.After decades of research and development, there are a lot of face detection algorithms and methods for face detection in which Adaboost method is one of the best. Discussion and in-depth analysis will be focus on this method for the first half of this article where the algorithm optimization, the practical problems encountered will be involved. Multi-gesture research and clustering method, as well as the SVM latter verifier module is introduced in this article. The major improvements include the following three points: 1) a variance of pixel were calculated so as to skip the regions of the picture which is either monotonous or over complex to speed up detection speed. 2) put forward a multi-profile prejudging and a new method of clustering. 3) the introduction of a SVM method using the wavelet extraction characters as a later verifier, reducing the false detection rate.The face feature point locating which is based on face detection is aimed to give the precise positions of all the face features. At present, the most successful method to solve this problem are ASM and AAM, etc. AAM, also known as dynamic apparent model, is first given out by F.T.Cootes in 1998 and is developed by S. Baker and others from Carnegie Mellon University's in 2001 who give the reverse calculation method of Hessian matrix in advance to reduce the iterative calculation. AAM is considered a effective way of facial feature point positioning method With three-dimensional scalability, and feature point positioning accuracy as well as up to 230 / s of processing speed, etc. However, this improved algorithm has the risk of illegal fitting which will lead to the wrong results. In the latter part of this paper, the AAM algorithm will be discussed and the probability damping method will be put forward to prevent the illegal fitting. This method can greatly improve the accuracy of fitting.