《An Introduction to Face Detection and Recognition：人脸检测和识别介绍》由会员分享，可在线阅读，更多相关《An Introduction to Face Detection and Recognition：人脸检测和识别介绍（43页珍藏版）》请在金锄头文库上搜索。
1、<p><p>&lt;p&gt;&amp;lt;p&amp;gt;&amp;amp;lt;p&amp;amp;gt;&amp;amp;amp;lt;p&amp;amp;amp;gt;An Introduction to Face Detection and Recognition Ziyou Xiong Dept. of Electrical and Computer Engineering, Univ. of Illinois
2、at Urbana-Champaign Outline nFace Detection nWhat is face detection? nImportance of face detection nCurrent state of research nDifferent approaches nOne example nFace Recognition nWhat is face recognition? nIts applications nDifferent approaches nOne example nA Video Demo What is Face Detection? nGi
3、ven an image, tell whether there is any human face, if there is, where is it(or where they are). Importance of Face Detection nThe first step for any automatic face recognition system system nFirst step in many Human Computer Interaction systems nExpression Recognition nCognitive State/Emotional Sta
4、te Recogntion nFirst step in many surveillance systems nTracking: Face is a highly non rigid object nA step towards Automatic Target Recognition(ATR) or generic object detection/recognition nVideo coding Face Detection: current state nState-of-the-art: nFront-view face detection can be done at 15 fr
5、ames per second on 320x240 black- and-white images on a 700MHz PC with 95% accuracy. nDetection of faces is faster than detection of edges! nSide view face detection remains to be difficult. Face Detection: challenges nOut-of-Plane Rotation: frontal, 45 degree, profile, upside down nPresence of bear
6、d, mustache, glasses etc nFacial Expressions nOcclusions by long hair, hand nIn-Plane Rotation nImage conditions: nSize nLighting condition nDistortion nNoise nCompression Different Approaches nKnowledge-based methods: nEncode what constitutes a typical face, e.g., the relationship between facial fe
7、atures nFeature invariant approaches: nAim to find structure features of a face that exist even when pose, viewpoint or lighting conditions vary nTemplate matching: nSeveral standard patterns stored to describe the face as a whole or the facial features separately nAppearance-based methods: nThe mod
8、els are learned from a set of training images that capture the representative variability of faces. Knowledge-Based Methods nTop Top-down approach: Represent a face using a set of human-coded rules Example: nThe center part of face has uniform intensity values nThe difference between the average int
9、ensity values of the center part and the upper part is significant nA face often appears with two eyes that are symmetric to each other, a nose and a mouth nUse these rules to guide the search process Knowledge-Based Method: Yang and Huang 94 nLevel 1 (lowest resolution): napply the rule “the center
10、 part of the face has 4 cells with a basically uniform intensity” to search for candidates nLevel 2: local histogram equalization followed by edge equalization followed by edge detection nLevel 3: search for eye and mouth features for validation Knowledge-based Methods: Summary nPros: nEasy to come
11、up with simple rules nBased on the coded rules, facial features in an input image are extracted first, and face candidates are identified nWork well for face localization in uncluttered background nCons: nDifficult to translate human knowledge into rules precisely: detailed rules fail to detect face
12、s and general rules may find many false positives nDifficult to extend this approach to detect faces in different poses: implausible to enumerate all the possible cases Feature-Based Methods nBottom-up approach: Detect facial features (eyes, nose, mouth, etc) first nFacial features: edge, intensity,
13、 shape, texture, color, etc nAim to detect invariant features nGroup features into candidates and verify them Feature-Based Methods: Summary nPros: Features are invariant to pose and orientation change nCons: nDifficult to locate facial features due to several corruption (illumination, noise, occlus
14、ion) nDifficult to detect features in complex background Template Matching Methods nStore a template nPredefined: based on edges or regions nDeformable: based on facial contours (e.g., Snakes) nTemplates are hand- coded (not learned) nUse correlation to locate faces Template-Based Methods: Summary n
15、Pros: nSimple nCons: nTemplates needs to be initialized near the face images nDifficult to enumerate templates for different poses (similar to knowledge- based methods) Appearance-Based Methods: Classifiers nNeural network nMultilayer Perceptrons nPrinciapl Component Analysis (PCA), Factor Analysis
16、nSupport vector machine (SVM) nMixture of PCA, Mixture of factor analyzers nDistribution Distribution-based method nNa&amp;amp;amp;amp;#239;ve Bayes classifier nHidden Markov model nSparse network of winnows (SNoW) nKullback relative information nInductive learning: C4.5 nAdaboost ? n? Face
17、and Non-Face Exemplars nPositive examples: nGet as much variation as possible nManually crop and normalize each face image into a standard size(e.g., 19&amp;amp;amp;amp;#215;19 nCreating virtual examples Poggio 94 nNegative examples: Fuzzy idea nAny images that do not contain faces nA large image subspace nBootstrapingSung and Poggio 94 Exhaustive Search nAcross scales nAcross locations Theory of Our Algorit&amp;amp;amp;lt;/p&amp;amp;amp;gt;&amp;amp;lt;/p&amp;amp;gt;&amp;lt;/p&amp;gt;&lt;/p&gt;</p></p>
