The Research On Human Face Detection In Video Monitoring System Based On Wavelet Analysis

Based on the characteristics of video monitoring system, according to the MPEG-4 standard, this thesis builds up a video coding model. And based on wavelet analysis technology, it researches on the human face detection algorithm within video sequence, and improves on the human face model of MPEG-4.Firstly, this thesis analyses the characteristics of video monitoring system, divides moving foreground video objects from the relatively static background according to the sprite coding. Following, a wavelet based face detection algorithm is presented for the moving regions (foreground video objects), and an amendment for face model within MPEG-4 is introduced. In addition, it adopts a multi-sampling filter series to decompose the sprite image and texture into wavelet coefficients which have the Multi-resolution Analysis (MRA) character. Finally, it adapts the distorted quantizing technology to realize the interest-based video coding, and realized the multi-band network application of video monitoring system.This thesis realized the real time face detection within video sequence primarily combined with skin model, movement detection and wavelet analysis technologies. At first, it pre-locates the face according to the human face model. Second, it premises that the face is moving and the background is relative static, hence, the face image can be located by the movement detection. Wavelet analysis has the significant features, e.g., localization of frequent-temporal and MRA. Thus, it is adopted to decrease the dimension of the face image and get the candidates of face with its wavelet coefficients. Finally, the face validation by the features of shape and gray information is carried out to pick out the real faces.Experiments show that this scheme is appropriate for the scenarios such as video monitoring system. It has solved the problem of real-time face detection within video sequence preferably and got well result.