The Multi-view Face Detection In Video Image

In recent years,face detection(Face Detection)has become a very active research topic in the field of computer vision and pattern recognition.With the development of intelligent monitor and human-computer interaction,people have an increasing requirement for ability of video processing,moving target auto-detection and an intelligent tract-monitoring video system.The system requires the face detection technology to overcome the technical problems,and then achieve the goal for multi-pose face detection and tracking.At present,face detection research mainly focuses on face detection with front view,and there are still some problems in effective methods of multi-pose face detection.This paper studies the multi-pose face detection theory and key technology of color video image.It analyzes the video images of the moving object extraction and studies face detection method based on skin color region segmentation of the model.The main work of the paper is as follows:(1)This paper analyzes the video image sequence types,and sums up the common methods of moving object detection.Then it puts forward the moving object segmentation method combining with background difference method and frame difference method.(2)Skin color is one of the important features of human faces.It does not depend on facial gesture and has the relative stability.This paper proposes the face detection method based on skin color model in YCbCr space.The algorithm first detects skin color region,according to the geometric characteristics of the human face,and then gets the full face region using mathematical morphology,with removing the non-facial parts including arms,legs,clothing and so on.(3)This paper analyzes the frontal face detection method and multi-pose face detection method,and it puts forward the multi-pose face detection method based on histogram and pyramid detector.Firstly,this detection technique uses a histogram to classify the multi-view training samples,and then it identifies the different angle and posture using the hierarchical model.The experimental results show that,compared with the classical algorithm,this method not only improves the recognition rate,but also reduces the computation time.