The Research Of Color Face Detection For Embedded System

The face recognition technology has matured and has been gradually applied to people's daily life in recent years. As the early stage of face recognition, face detection affects the speed and recognition rate to some extent. Therefore , the study of face detection algorithm to improve the face detection rate and detection rate, can improve the overall performance of face recognition system.Firstly, in this paper, skin-color pretreatment should be done to people's faces which are going to be detected. By analyzing the skin-clusters in different color space, establish skin-color model among the YCbCr color space. Less precision needed by the skin-color pretreatment, the skin color models established in this paper are the most simple and fastest threshold segmentation model. The model provides a range of the threshold. If pixels in the Cb and Cr components belong to the threshold range, the pixels are determined to be skin-color pixel. Otherwise, they are judged as non-skin pixel. Skin-color matrix can be generated by using skin-color pretreatment algorithm. The skin-color matrix is a binary one, where 1 represents the skin-color pixel and -1 represents non-skin pixel.Secondly, improve AdaBoost-based face detection algorithm. In the end of the classifier training algorithm, the false positive of the weak classifier may be close to 50% so that it formed a joint strong classifier performance is not significantly improved. As a result , the algorithm reduces the classification performance and decreases the detection rate.The reasons why they happen are as follows: 90 percent of the edges are about the same in gray value and in color image.There are still 10 percent of the edges left may not be detected in intensity images. Training in the late, classified information contained gray feature can not distinguish human face samples with non-face ones. In this paper points graph of color concepts and puts the color information into AdaBoost training algorithm to generate the classifiers corresponding the three color components CY(x)C_Cb(x)and C_Cr(x). Through improving the structure of cascade classifier, the three classifiers improve detection rate of the algorithm. Because Cb and Cr color components wipe off the influence of the brightness, it also increases the adaptability of the algorithm to different light and colors.Finally, algorithm of skin color pretreatment and improved AdaBoost's face detections are applied to embedded human-machine interface systems.With skin-color matrix ruling out most of the non-skin areas, AdaBoost classifier can focus more on possible facial skin areas, thereby enhance the detection algorithm speed and detection rate.