彩色图像中的人脸检测外文文献翻译英文原文.docVIP

696 IEEETRANSACTIONSONPATTERNANALYSISANDMACHINEINTELLIGENCE, VOL.24, NO.5, MAY2002 FaceDetectioninColorImages beliefnetworks[46]provides facedetection fornonfrontal views. Geometrical facial templates and the Hough transform were incorporated to detect gray-scale frontal faces in real time applications [24]. Face detectors based on Markov random fields [23]andMarkovchains[8],makeuseofthespatialarrangementof pixelgrayvalues. Rein-Lien Hsu, Student Member,IEEE, Mohamed Abdel-Mottaleb, Member,IEEE ,and Anil K. Jain, Fellow,IEEE Categorizing face detection methods based on the representa- tion used reveals that detection algorithms using holistic repre- sentations have the advantage of finding small faces or faces in poor-qualityimages,whilethoseusinggeometricalfacialfeatures provide a good solution for detecting faces in different poses. A combinationofholisticandfeature-basedapproaches[13],[21]isa promising approach to face detection as well as face recognition. Motion [9], [11] and skin-tone color [11], [4], [38], [43], [12] are useful cues for face detection. However, the color-based ap- proaches face difficulties in robustly detecting skin colors in the presenceofcomplexbackgroundanddifferentlightingconditions [12].Weproposeafacedetectionalgorithmthatisabletohandlea widerangeofvariationsinstaticcolorimages,basedonalighting compensationtechniqueandanonlinearcolortransformation.Our approach models skin color using a parametric ellipse in a two- dimensionaltransformedcolorspaceandextractsfacialfeaturesby constructingfeaturemapsfortheeyes,mouth,andfaceboundary. Section2describesthefacedetectionalgorithm.Section3presents the detection results of our algorithm on several face databases. ConclusionsandfutureworkaredescribedinSection4. AbstractDHumanfacedetectionplaysanimportantroleinapplicationssuchas videosurveillance,humancomputerinterface,facerecognition,andfaceimage databasemanagement.Weproposeafacedetectionalgorithmforcolorimagesin thepresenceofvaryingli