| Face is a kind of natural structure with complex intricacies change, the test results vulnerable to light, the impact on the complexity of the background and noise present in the image, which increased the difficulty of face detection. Currently, a new face detection algorithm are emerging, but most of them are for general background or specific constraints, face detection for complex background image still need further research to be done.Skin detection has these characteristics that robustness to gesture, facial expression and detection speed is quick as well as false detection rate is high. The adaboost detection algorithm whose detection rate is high and the number of training sample is large what’s more the training time is long. According to the characteristics of the two algorithms, the paper proposed an algorithm that combined skin detection and variance feature of haar, and used SVM instead of adaboost for training classifier, to improve detection accuracy while reducing the detection time.First transformed YCbCr color space to YCb’ Cr’ space by nonlinear transformation, in which color space established elliptical skin model and used logistic regression analysis to determine the probability of every bit of skin color, generated the skin probability diagram, which mapped every bit of pixel values to [0,1], and then on the basis of the Ostu method, parallel genetic algorithm was used to determine the threshold of skin color segmentation, With which binary the image. The mentality of image segmentation was performed after chemical treatment to remove small particles noise. In order to improve the accuracy of detection while reducing the detection time, this paper extracted Haar features from human face image, and integral image was used to calculate the value. To reduce the number of features, the paper instead rectangle feature with a small amount of Haar variance characteristics, and combined the SVM training classification method to authenticate the face region. |
PDF Full Text Request