Research And Implementation On Face Detection Based On Improved AdaBoost

Face detection is one of the important areas in detection of biological characteristics.It has broad prospects in application for the merit of non-contact acquisition and non-invasive,and AdaBoost is the main algorithm of it.This paper focus on the AdaBoost algorithm and make three improvements to raise the detection speed and detection rate in weak illumination.The detailed contends is as follows:Firstly,this paper detects the skin-color area in YCrCb space,which will decrease the search area of AdaBoost.This paper detect skin-color area in YCrCb space as it can be segmentated better through the value of Cr and Cb in this space.The distribution of vaule of Cr and Cb should be fixed through experiments before segmentating.And then in segmentation,every pixel's value of Cr and Cb will be tested and classified in YCrCb space.The segmentated pixels make up the skin-color area,which will be detected by AdaBoost.While in the experiments,a serious problem appears when detecting weak-light images,that is:the skin-color area is detect wrongly or even missed.Secondly,the method of HE-lgDCT is used to compensate the illumination,which is designed to solve the problem of false and missed detect in weak-light image.The Y(illumination) will be detect after inputting an image.If lots of Y focus on low-value area,the image will need illumination compensation:HE(histogram equalization) to gray image and lgDCT(Discrete Cosine Transform in log domain) for RGB image.Illumination compensation increases the detection rate of images in weak-light and decreases false and missed detection,though it takes a little more time.Thirdly,Gabor features replace the Harr-like features to achieve better face expression result.And then PCA is used to reduce the Gabor-features dimension to achieve faster detection speed.While there are amounts of Gabor featues,so it need PCA to reduce the feature-dimension.PCA not only minimizes the mean-square-error of Gabor features,but also performs well in expressing faces in low dimension.After dimension reduction with PCA,the paper chooses fourteen Gabor features with high contribution rate and then train them to weak classifiers respectively,compose them to a strong classifier in AdaBoost at last.At last,the papaer builds up a Face Detecting System,which contains three improvements as mentioned above.And the test shows that the system increases 39 percent detection speed and 23 percent detection rate in weak light,at the same time,it meets the requiments of real-time detection.