li, jain - handbook od face recognitiion

[...]... for other face detection methods 1 Appearance-Based and Learning Based Approaches With appearance-based methods, face detection is treated as a problem of classifying each scanned subwindow as one of two classes (i.e., face and nonface) Appearance-based methods avoid difficulties in modeling 3D structures of faces by considering possible face appearances under various conditions A face/ nonface classifier... locations and scales by a subwindow Face detection is posed as classifying the pattern in the subwindow as either face or nonface The face/ nonface classifier is learned from face and nonface training examples using statistical learning methods This chapter focuses on appearance-based and learning-based methods More attention is paid to AdaBoost learning-based methods because so far they are the most... the face pattern and classification methods used to distinguish between faces whereas face localization and normalization are the basis for extracting effective features These problems may be analyzed from the viewpoint of face subspaces or manifolds, as follows Image/Video Face Location, Face Size & Pose Detection -Tracking Feature Vector Aligned Face Face Alignment Feature Extraction Face. .. distribution for face, linear dimension reduction is performed to obtain the PCA modes The likelihood density is estimated using PCA and its residuals, making use of Bayesian techniques [25] The nonface class is modeled similarly A classification decision of face/ nonface is made based on the two density estimates The BDF classifier is reported to achieve results that compare favorably with state-of-the-art face. .. nonfrontal faces is important for many real applications because approximately 75% of the faces in home photos are nonfrontal [17] A reasonable treatment for the multiview face detection problem is the view-based method [29], in which several face models are built, each describing faces in a certain view range This way, explicit 3D face modeling is avoided Feraud et al [7] adopt the view-based representation... with out-of-plane (left-right) rotation A coarse-to-fine, simple-tocomplex architecture, called a detector-pyramid, is designed for the fast detection of multiview faces This work leads to the first realtime multiview face detection system It runs at 200 ms per image (320×240 pixels) on a Pentium-III CPU of 700 MHz Lienhart et al [22] use an extended set of rotated Haar features for dealing with in-plane... rotation: (1) out-of-plane (left-right) rotation; (2) in-plane rotation; and (3) up-and-down nodding rotation Adopting a coarse-to-fine view-partition strategy, the detector-pyramid architecture consists of several levels from the coarse top level to the fine bottom level Rowley et al [31] propose to use two neural network classifiers for detection of frontal faces subject to in-plane rotation The first is... in-plane rotation Left and right: In-plane rotated by ±30◦ The design of the detector-pyramid adopts the coarse-to-fine and simple-to-complex strategy [2, 8] The architecture is illustrated in Figure 2.12 This architecture design is for the detection of faces subject to out-of-plane rotation in Θ = [−90◦ , +90◦ ] and in-plane rotation in Φ2 = [−15◦ , +15◦ ] The full in-plane rotation in Φ = [−45◦ , +45◦... simple color-based face detection algorithm consists of two steps: (1) segmentation of likely face regions and (2) region merging Fig 2.2 Skin color filtering Input image (left) and skin color-filtered map (right) A skin color likelihood model, p(color|f ace), can be derived from skin color samples This may be done in the hue-saturation-value (HSV) color space or in the normalized red-green-blue (RGB)... [−45◦ , +45◦ ] is dealt with by applying the detector-pyramid on the images rotated ±30◦ , as mentioned earlier Fig 2.12 Detector-pyramid for multiview face detection Coarse-to-fine The partitions of the out-of-plane rotation for the three-level detector-pyramid is illustrated in Figure 2.13 As the the level goes from coarse to fine, the full range Θ of out-ofplane rotation is partitioned into increasingly . 0-3 8 7-4 0595-X (alk. paper) 1. Human face recognition (Computer science I. Li, S. Z., 1958– II. Jain, Anil K., 1948– TA1650.H36 2004 006.4′2—dc22 2004052453 ISBN 0-3 8 7-4 0595-X Printed on acid-free. algorithms and other factors that af- fect face detection performance. Chapters 3 and 4 discuss face modeling methods for face alignment. These chapters de- scribe methods for localizing facial components. 4882 4-1 226 Beijing 100080 USA China jain@ cse.msu.edu szli@nlpr.ia.ac.cn Library of Congress Cataloging-in-Publication Data Handbook of face recognition / editors, Stan Z. Li & Anil K. Jain. p.