【正文】 一張人臉 檢出1 漏檢0 虛警2 一張有遮擋的人臉 檢出1 漏檢0 虛警0 四張有遮擋的人臉 檢出4 漏檢0 虛警2 五張人臉 檢出5 漏檢0 虛警0 五張遠(yuǎn)近不同的人臉 檢出5 漏檢0 虛警3 三張人群中模糊的人臉 檢出3 漏檢0 虛警1 八張人臉 檢出8 漏檢0 虛警1 基于AdaBoost算法的視頻檢測(cè)結(jié)果 一張人臉 檢出1 漏檢0 虛警0 四張人臉 檢出3 漏檢1 虛警2 光線昏暗且有遮擋的兩張人臉 檢出2 漏檢0 虛警17 結(jié)論與對(duì)未來(lái)的展望通過(guò)對(duì)上述測(cè)試結(jié)果的分析，我們得知一個(gè)完備的人臉檢測(cè)系統(tǒng)是很難被設(shè)計(jì)出來(lái)的，特別是將單一系統(tǒng)應(yīng)用于不同領(lǐng)域內(nèi)的人量檢測(cè)。在本文中，我們對(duì)AdaBoost算法進(jìn)行了綜合的闡述，并對(duì)haar特征進(jìn)行了擴(kuò)展，使其更接近于一個(gè)完備的特征集。當(dāng)然可以擴(kuò)展的地方還有很多，比如近幾年來(lái)有很多前輩在AdaBoost算法本身和強(qiáng)分類器級(jí)聯(lián)結(jié)構(gòu)方面的改進(jìn)也做出了很大的貢獻(xiàn)。在這里我們對(duì)未來(lái)的人臉檢測(cè)工作做出一些展望：我們可以嘗試引進(jìn)更多類型的特征，如三角特征，并將三角特征與矩形特征結(jié)合使用。同時(shí)，對(duì)于不同人臉檢測(cè)方法的融合方面，我們也有很多工作可以做，畢竟對(duì)于不同領(lǐng)域，每一人臉檢測(cè)算法都有其長(zhǎng)足之處。 參考文獻(xiàn)[1] Belhumeur ., Hespanha ., Kriegman . 1997. Eigenface vs. Fisherface: Recognition Using Class Specific Linear Projection. IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 19(7), 711~720[2] E. Osuna, R. Freund, and F. Girosi. 1997. Training Support Vector Machines: An Application to Face Detection. Proc. IEEE Conf. Computer Vision and Pattern Recognition, 130~136[3] Huang C, Ai H Z, Li Y, et al. 2007. High performance rotation invariant multiview face detection. IEEE Trans. PAMI, 29(4), 671~686[4] Viola P, Jones M. 2001. Rapid object detection using a boosted cascade of simple , 511~518[5] Jia H X, Zhang Y J. 2007. Fast human detection by boosting histograms of oriented , 683~688[6] [7] Rowley . Neural networkbased face Thesis, CMU[8] 章毓晉. 2006. 圖像處理（上冊(cè)）：圖像處理. 北京：~201[9] Liu C, Shum H Y. 2003. KullbackLeibler boosting. Proc. CVPR, 587594[10] Dalal N. 2006. Finding people in images and videos. Doctoral Dissertation. Institut National Polytechnique Grenoble, France.[11] Comaniciu algorithm for datadriven bandwidth selection. IEEE Trans. PAMI, 25（2）：281~288[12] 趙楠. 2005. 基于AdaBoost算法的人臉檢測(cè). 學(xué)士學(xué)位論文. 北京大學(xué)[13] Lienhart R, Maydt . An extended set of Haarlike features for rapid object detection. Proc. ICIP, 900~903[14] ChungChih . A New Feature Set for Face Detection. Master Thesis, National Tsing Hua University[15] Viola P., Jones M. . Robust RealTime Face Detection. International Journal of Computer Vision 57（2）, 137~154[16] 章毓晉等. 2009. 基于子空間的人臉識(shí)別. 北京：清華大學(xué)出版社. 1~46, 164~183[17] 于仕琪，張兆翔. 2010. OpenCV中文參考手冊(cè). . 1~211[18] 譚浩強(qiáng). 2009. C程序設(shè)計(jì)（第三版）. 北京：清華大學(xué)出版社. 10~139[19] 嚴(yán)蔚敏 吳偉民. 2008. 數(shù)據(jù)結(jié)構(gòu)（C語(yǔ)言版）. 北京：清華大學(xué)出版社. 69~105[20] Rafael , Richard . Digital Image Process(Second Edition).Prentice Hall. 460~514[21] 張兆禮，趙春暉，梅曉丹. 2001. 現(xiàn)代圖像處理技術(shù)及Matlab實(shí)現(xiàn). 北京：人民郵電出版社. 83~142 英文原文2010 Second International Conference on Computer Modeling and SimulationAdaBoostbased Face Detection in Color Images with Low False AlarmSania Arjomand InalouDepartment of Electrical EngineeringInternational Sharif University of TechnologyKish Island, IranShohreh KasaeiDepartment of Computer EngineeringSharif University of TechnologyTehran, IranskasaeiAbstractIn this paper, we have proposed a new face detection method which bines the AdaBoost algorithm with skin color information and support vector machine (SVM). First, a cascade classifier based on AdaBoost is used to detect faces in images. Due to noise and illumination changes some nonfaces might be detected too, therefore we have used a skin color model in the YCbCr color space to remove some of the detected nonfaces. Finally, we have utilized SVM to detect faces more accurately. Experimental results show that the performance of the proposed method is higher than the basic AdaBoost in the sense of detecting fewer nonfaces.Keywordsface detection。 skin color features。 such as RGB, HSV, YCbCr, YUV, and CIELab. However, it has always been a disputable issue that which color space is the best for describing the property of human skin. The next step of skin color detection is to build a decision rule that discriminates between skin and nonskin pixels. This is usually acplished by the skin color modeling method. Current approaches can be classified into two main classes. (i) parametric methods (such as the single Gaussian, and mixtures of Gaussians), and (ii) nonparametric methods (such as Bayes classifier and self organizing map (SOM)). Furthermore some new approaches are based on learning (such as cellular learning automata (CLA)).C. Support Vector MachineSupport vector classifiers implicitly map the data into a high dimensional feature space via a nonlinear transform and pute a hyperplane which separates the data in the feature space by a large margin. Intuitively, a good choice is a hyperplane that leaves the maximum margin between the two classes (–1, +1) and minimizes a quantity proportional to the number of misclassification errors. Its dual quadratic programming classification problem is: （6）Where are the Lagrange multipliers, and C is a constant variable that controls the tradeoff between misclassification error and the margin.The functional form of the mappingdoes not need to be known since it is defined by the choice of kernel （7）The vector w has the form of and therefore （8）The training examples with are called support vectors.III. PROPOSED FACE DETECTION METHODIn this paper, we have proposed a face detection method for color images using AdaBoost algorithm bined with skin color information and SVM. Fig. 3 shows the overall structure of the proposed method.In the first stage, like Viola and Jones, we have used a cascade classifier which is constructed by AdaBoost algorithm to detect faces. As we mentioned above, there are three main contributions of the AdaBoostbased face detection. The first is the integral image which allows the Haarlike features to be puted very rapidly. The second is the AdaBoost algorithm which selects a small number of Haarlike features from a larger set and creates efficient classifiers. The third contribution is a method for bining classifiers in a cascade in order to improve the putational efficiency and to also reduce the false positi