Y. Yacoob and L.S. Davis. Computing spatio-temporal representations of human faces. In Proc. Computer Vision and Pattern Recognition, CVPR-94, pages 70#75, Seattle, WA, June 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....on groups of people from their facial expressions. Black and Yacoob [3] describe a system that recognizes facial expressions in presence of signi cant head motions. They use parameterized optical ow models to track rigid and non rigid facial movements. In an earlier version Yacoob and Davis [34] use optical ow at high gradient points on the face to recognize facial expressions. Essa and Pentland [18] analyze the facial expressions using optical ow in an estimation and control framework coupled with a physical model describing the skin and muscle structure of face. Zhang [37] has ....

Y. Yacoob and L. Davis. Computing spatio-temporal representation of human faces. In CVPR, pages 70-75, Seattle, WA, June 1994.

....the flow can be more accurately represented using a low order basis. At one extreme are methods which use models of the facial muscles [6] A variety of less constrained flexible models have been studied in the context of face recognition [9, 17, 19, 11] At the other extreme, Yacoob and Davis [18] use very little modeling, segmenting the face and developing statistical descriptions of the changes of the segmented regions. Black and Yacoob [3] used mid level representations of flow over small image regions. This work is related to ours, in that they model flow using the lowest order Zernike ....

Y. Yacoob and L. Davis. Computing spatio-temporal representations of human faces. In CVPR, pages 70--75, 1994.

....applications call for an unrestricted and robust head tracking system from 2D monocular image sequences. 1.1. Previous Work In recent years considerable progress has been made on the problem of head face tracking from 2D monocular image sequences. Some systems extract the 2D position of the head [9, 11], while others retrieve the 3D motion parameters [6, 8, 5, 3, 1, 4, 7, 6, 2, 10] In this paper we concentrate on 3D head tracking. Li et al. 8] used an affine model to describe both rigid and non rigid facial motion. Their approach was characterized by a render feedback loop connecting computer ....

Y. Yacoob and L.S. Davis. Computing spatio-temporal representations of human faces. In CVPR94, pages 70--75, 1994.

....rigid motion tracking using unit normal changes and non rigid motion tracking using discriminant changes. 2.1 Introduction Our study is based on a sequence of 3D face data. This kind of data set is different from other existing facial recognition studies, where 2D intensity data is used [20, 27, 28]. There are some advantages of using 3D range data: i) more geometric hints for facial feature points detection; ii) facial parameters are 3D oriented; iii) 3D movements constitute a large part of facial gestures. An example is FAP#14 (thrust jaw) which is defined as the depth displacement of jaw. ....

Y. Yacoob and L. Davis, Computing spatio-temporal representations of human faces, Computer Vision and Pattern Recognition Conference, 1994

....and Pentland [JP97] to track head motion using a small number of image features. The rough 3 D shape of the head is also extracted. Another approach is to directly use the optical flow field from face images. Yacoob and Davis use statistical properties of the flow for expression recognition [YD94] Black and Yacoob parameterize the flow field based on the structure of the face under projection [BY95] Basu, et al. [BEP96] extract a flow field, and then regularize it using a 3 D ellipsoid model of the head. Addressing the problem of image coding, Li, et al. [LRF93] estimate face motion using ....

....7.1(c) The distance from the initial face to the camera is determined given the assumption that the subject s face is the same size as the model. a) b) c) Figure 7. 1: Model initialization The problem of automatically locating the face and its various features has been addressed elsewhere [YD94, YCH92] and could be used to make this process automatic. No markers or make up are used on the subject (markers are used for the validation of the tracking method, however, as described below) Experience has shown that the initialization process is robust to small displacements (i.e. several ....

Y. Yacoob and L.S. Davis. Computing spatio-temporal representations of human faces. In Proceedings CVPR '94, pages 70--75, 1994.

....the Smart Rooms [50] and KidsRoom [8] systems from the MIT Media Laboratory. Other instances of smart and interactive environments that use computer vision techniques to track and recognize human activity are being developed at the MIT AI Laboratory [43, 42] CMU [29, 57] University of Maryland [27, 71] and several other leading research institutions. 2.5 Research Training Activities and Education Involving The System We expect the development of this equipment will involve a substantial number of graduate and undergraduate students, and be the subject of many PhD and undergraduate theses. ....

Y. Yacoob and L. Davis. Computing spatio-temporal representations of human faces. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 70--75. IEEE Computer Society, 1994. 20

....Thrun and Thorpe [12] formulated the 3D tracking of planar patches using texture mapping as the measurement model in an extended Kalman filter framework. Several other techniques have been proposed for free head motion and face tracking. Some of these techniques focus on 2D tracking (e.g. [4, 9, 14, 16, 27, 35, 36]) while others focus on 3D tracking or stabilization. Some methods for recovering 3D head parameters are based on tracking of salient points, features, or 2D image patches. The outputs of these 2D trackers can be processed by an extended Kalman filter to recover 3D structure, focal length and ....

Y. Yacoob and L.S. Davis. Computing spatio-temporal representations of human faces. IEEE Trans. on Patt. Analysis and Machine Intelligence, 18(6):636--642, 1996.

....in orientation, or scale. Moreover, they must work near video frame rates. Such requirements make the problem even more challenging. 1.1 Previous Work In recent years several techniques have been proposed for 3D head motion and face tracking. Some of these techniques focus on 2D tracking (e.g. [6, 10, 16, 22, 23]) while others focus on 3D tracking or stabilization. Some methods for recovering 3D head parameters are based on tracking of salient points, features, or 2D image patches [1, 12] Others use optic flow to constrain the motion of a rigid or non rigid 3D surface model [2, 7] In [14] a ....

Y. Yacoob and L.S. Davis. Computing spatio-temporal representations of human faces. PAMI, 18(6):636--642, 1996.

....then, motion based recognition has become an active area of study within computer vision. These efforts include identification of pedestrians [6, 7] MLD based motion recognition 4 STEVEN M. SEITZ AND CHARLES R. DYER [8] hand gesture recognition [9, 10] interpretation of facial expressions [11, 12], and temporal textures [13] Details of many of these efforts are described in other chapters of this book, and in a recent survey article [14] Cyclic motion analysis is unique among motion based recognition approaches in that it does not require any type of object or motion specific model. ....

....of many of these efforts are described in other chapters of this book, and in a recent survey article [14] Cyclic motion analysis is unique among motion based recognition approaches in that it does not require any type of object or motion specific model. Other motion based recognition techniques [6, 7, 9, 10, 11, 12, 13] require a priori models of the underlying object and or the motion in the scene, although these models can potentially be learned [9, 15] In contrast, periodicity is a universal motion characteristic that can be detected and described without knowledge of the underlying object and without ....

Y. Yacoob and L. Davis, "Computing spatio-temporal representations of human faces," in Proc. Computer Vision and Pattern Recognition Conf., pp. 70--75, 1994.

....Analysis A final area of relevance to FRT is the motion analysis of non rigid objects [117, 118, 119, 120] Some of the work [121, 122] is potentially useful in face recognition. Another application of non rigid motion to faces to is the recognition of facial expressions from image sequences [123]. 4.2 Tracking, modeling and non face based recognition During the past five years, tracking, modeling and recognition of hand gestures and human behaviors have been extensively studied. We briefly review some of these topics here. Research on human emotion recognition has been extended to a new ....

....of the facial features on the interpretation of facial expressions. Bassili [134] suggested that motion in the image of a face could allow emotions to be identified even with minimal information about the spatial arrangement of the features. In the engineering literature, early efforts [123, 135] were based on analysis of the optical flow field of the image sequence, which provides clues to the spatial changes in the facial features. 128] demonstrated successful facial expression recognition in extensive laboratory experiments involving 40 subjects as well as in television and movie ....

Y. Yacoob and L. S. Davis, "Computing Spatio-Temporal Representations of Human Faces," in Proceedings, IEEE Conference on Computer Vision and Pattern Recognition, 1994.

....Almost all of this work treats face recognition as a static problem approached by pattern recognition techniques 1 applied to single static images. Only recently the attention of researchers has shifted to the temporal aspect of facial expressions by using optical flow in sequences of face images [41, 42, 25, 65]. We do not intend to give a comprehensive overview of face recognition here. Rather, we will summarize some ideas that are relevant to our work. Recently, a systematic comparison of typical approaches (feature based versus template based techniques) to face recognition was carried out by ....

Y. Yacoob and L. S. Davis. Computing spatiotemporal representations of human faces. In IEEE Proc. of CVPR, Seattle, WA, June 1994. 19 A Hierarchical estimation of global pose from local displacements We assume an affine model for the displacement vector field. The affine displacement field d(x i ) is determined at

....that these restrictions have begun to be lifted by looking at multiple view based systems and flexible matching procedures. Outside of the recognition problem, there have been recent studies on analyzing faces under different lightings (Hallinan [64] and expressions (Essa [51] Yacoob and Davis [142], Beymer, Shashua, and Poggio [19] While the prototypical face recognition system deals with intensity images of frontal or near frontal views, there are systems that are based on 3D range measurements and others that utilize the facial profile seen in side views of the face. More will be said ....

Yaser Yacoob and Larry Davis. Computing spatio-temporal representations of human faces. In Proceedings IEEE Conf. on Computer Vision and Pattern Recognition, pages 70--75, Seattle, WA, 1994.

....the flow can be more accurately represented using a low order basis. At one extreme are methods which use models of the facial muscles [6] A variety of less constrained flexible models have been studied in the context of face recognition [9, 17, 19, 11] At the other extreme, Yacoob and Davis [18] use very little modeling, segmenting the face and developing statistical descriptions of the changes of the segmented regions. Black and Yacoob [3] used mid level representations of flow over small image regions. This work is related to ours, in that they model flow using the lowest order Zernike ....

Y. Yacoob and L. Davis. Computing spatio-temporal representations of human faces. In CVPR, pages 70--75, 1994.

....automatic and required the facial features to be highlighted with special make up. Later techniques have overcome this limitation [49, 50] Most research on facial expression recognition is based only on nonrigid facial deformation patterns. Techniques used for motion analysis are optical flow [51, 52], 2D graphical models ( potential nets ) 53] and local parametric models [54, 55] Appearance variations due to facial expressions that are not well described by motion fields are ignored. Most expression recognition algorithms were applied solely to the apex of the facial deformation, ....

Y. Yacoob and L. Davis, "Computing spatio-temporal representations of human faces," in IEEE Int. Conf. on Computer Vision, Cambridge, MA, IEEE Computer Society Press, pp. 70--75, 1995.

No context found.

Y. Yacoob and L.S. Davis. Computing spatio-temporal representations of human faces. In Proc. Computer Vision and Pattern Recognition, CVPR-94, pages 70#75, Seattle, WA, June 1994.

No context found.

Y. Yacoob and L.S. Davis. Computing spatio-temporal representations of human faces. In Proc. Computer Vision and Pattern Recognition, CVPR-94, pages 70--75, Seattle, WA, June 1994.

....attention as a motion estimation problem. Previous work has typically focused on one part of the problem or the other: either rigid head tracking [1] with no facial expressions or expression recognition with either no motion at all [10] or a roughly stationary head with a changing expression [7, 8]. Models used in recognizing facial expressions vary in the amount of information about head shape and motion they contain. At one extreme are approaches which employ physically based models of heads including skin and musculature [6, 7] Slightly weaker models use deformable templates to ....

....7] Slightly weaker models use deformable templates to represent feature shapes in the image [10] Approaches that determine the expression by matching stored image templates to the current image [5] use even less explicit spatial information. At the other extreme is the work of Yacoob and Davis [8] in which facial expressions are recognized in image sequences using statistical properties of the optical flow with only very weak models of facial shape. In this paper we explore a middle ground between the template based approaches and the optical flow based approaches. The piecewise parametric ....

[Article contains additional citation context not shown here]

Y. Yacoob and L.S. Davis. Computing spatio-temporal representations of human faces. In CVPR-94, pp. 70-- 75, Seattle, WA, June 1994.

No context found.

Y. Yacoob and L. Davis. Computing spatio-temporal representations of human faces. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 70--75. IEEE Computer Society, 1994.

No context found.

Y. Yacoob and L. Davis. Computing spatio-temporal representations of human faces. In CVPR'94, pages 70--75, 1994.

No context found.

Y. Yacoob and L. S. Davis, "Computing spatio-temporal representations of human faces," in Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (IEEE Computer Society, Los Alamitos, Calif., 1994), pp. 70--75.

No context found.

Yacoob Y, Davis L. Computing spatio-temporal representations of human faces. In: Computer Vision and Pattern Recognition Proceedings. IEEE Computer Society 1994

No context found.

Y. Yacoob and L. Davis, \Computing spatiotemporal representations of human faces," in IEEE International Conference on Computer Vision, (Cambridge, MA), IEEE Computer Society Press, June 1995, pp. 70-75.

No context found.

Y. Yacoob and L. S. Davis. Computing Spatio-temporal Representations of Human Faces. In Proceeding of CVPR, pages 70--75, 1994.

No context found.

Yaser Yacoob and Larry Davis. Computing Spatio-Temporal Representations of Human Faces. Technical report, Computer Vision Labratory, University of Maryland, 1994.

No context found.

Y. Yacoob, L. Davis, "Computing Spatio-Temporal Representations of Human Faces," CVPR, pp. 70-75, 1994.

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC