Study On Multi-pose Face Detection

In recent years, due to the increasingly important status of the face detectiontechnology in facial image processing and analysis, face detection has graduallydeveloped into a relatively independent research field and a leading direction, which getsmore and more extensive attention. Although, people in the research of face detectionhave made some achievements, however, in the actual demand there are still a lot ofproblems. This paper is to solve the existing problems in multi-pose face detection forfurther study.With the development of the face detection technology, detection rates of the positivefacial image have achieved satisfactory results. However, for the facial images whichexists multi-pose, many popular methods are not applicable. The most common method,face detection based on characteristics, has not too much limitation to the facial imageswhich exists multi-pose, which can be used to solve the multi-pose face detection. But theface detection methods based on characteristic exists a problem that is its detection rate isnot high.At present, the face detection based on Adaboost algorithm is a kind of face detectionmethod with high detection rate, but because as the face feature its Haar-like feature isgenerally designed for positive facial images, so the effect for this traditional facedetection method is not good.The methods based on the characteristics for facial image posture have less limits, so,combining the face detection based on skin-color characteristics with the edge detectionmethod, we get a preliminary inspection and adjust the images to positive facial image.We can introduce several new rectangular characteristics to the traditional face detectionbased on Adaboost algorithm, in order to improve the detection rate. Put the positive faceimage which got in preliminary test into the new characteristics face detection. Comparedwith the traditional method, this paper introduces new characteristics and how to getpositive facial image, these methods can effectively improve the detection rate.