M. Jones and T. Poggio. Multidimensional morphable models: A framework for representing and matching object classes. In Proceedings of the IEEE International Conference on Computer Vision, pages 683--688, 1998.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....gradient descent, inverse compositional image alignment. 1 Introduction Active Appearance Models (AAMs) Cootes et al. 2001] first proposed in [Cootes et al. 1998] and the closely related concepts of Active Blobs [Sclaroff and Isidoro, 1998] and Morphable Models [Vetter and Poggio, 1997, Jones and Poggio, 1998, Blanz and Vetter, 1999] are non linear, generative, and parametric models of a certain visual phenomenon. The most frequent application of AAMs to date has been face modeling [Lanitis et al. 1997] However, AAMs may be useful for other phenomena too [Sclaroff and Isidoro, 1998, Jones and ....

....and Poggio, 1998, Blanz and Vetter, 1999] are non linear, generative, and parametric models of a certain visual phenomenon. The most frequent application of AAMs to date has been face modeling [Lanitis et al. 1997] However, AAMs may be useful for other phenomena too [Sclaroff and Isidoro, 1998, Jones and Poggio, 1998] Typically an AAM is first fit to an image of a face; i.e. the model parameters are found to maximize the match between the model instance and the input image. The model parameters are then used in whatever the application is. For example, the parameters could be passed to a classifier to ....

[Article contains additional citation context not shown here]

M. Jones and T. Poggio. Multidimensional morphable models: A framework for representing and matching object classes. In Proc. ICCV, pages 683--688, 1998.

....Invariance, Face Recognition, Segmentation. I. Introduction M ODELS capable of synthesising complete images of objects have over the past few years proven their worth when interpreting unseen images. Applications include real time tracking of deformable objects [1] 2] face recognition [3] [4], 5] and recovery of anatomical structures in magnetic resonance images [6] 7] 8] 9] ultrasound images [10] and x rays [9] 11] The key idea to all of these generative models is to perform a per pixel comparison between unseen input images and synthesised images and subsequently drive ....

M.J. Jones and T. Poggio, \Multidimensional morphable models: a framework for representing and matching object classes," International Journal of Computer Vision, vol. 29, no. 2, pp. 107-31, 1998.

....faces were developed for face recognition and image coding [4, 18, 37] Different approaches have been taken to automate the matching step necessary for building up morphable models. One class of techniques is based on optic flow algorithms [5, 4] and another on an active model matching strategy [12, 16]. Combinations of both techniques have been applied to the problem of image matching [36] In this paper we extend this approach to the problem of matching 3D faces. The correspondence problem between different threedimensional face data has been addressed previously by Lee et al. 20] Their ....

....of the mesh would be redundant. In each iteration, we therefore select a random subset K#f1; n t g of 40 triangles k and replace EI by EK = X k2K kIinput##p x;k ; # p y;k # , Imodel;k #k 2 : 7) The probability of selecting k is p#k 2K# # ak . This method of stochastic gradient descent [16] is not only more efficient computationally, but also helps to avoid local minima by adding noise to the gradient estimate. Before the first iteration, and once every 1000 steps, the algorithm computes the full 3D shape of the current model, and 2D positions #px;p y # T of all vertices. It ....

M. Jones and T. Poggio. Multidimensional morphable models: A framework for representing and matching object classes. In Proceedings of the Sixth International Conference on Computer Vision, Bombay, India, 1998.

....models. In the following we describe the techniques in more detail and give examples of the model, its ability to estimate pose, to track faces and to synthesize unseen views. 12.1 Related Work Statistical models of shape and texture have been widely used for recognition, tracking and synthesis [39, 47, 22, 70], but have tended to only be used with near fronto parallel images. Moghaddam and Pentland [52] describe using view based eigenface models to represent a wide variety of viewpoints. Our work is similar to this, but by including shape variation (rather than the rigid eigen patches) we require ....

M. J. Jones and T. Poggio. Multidimensional morphable models : A framework for representing and matching object classes. International Journal of Computer Vision, 2(29):107-131, 1998.

....that the models can be used to track faces through wide angle changes, and that they can be used to predict appearance from new viewpoints given a single image of a person. 11.8 Related Work Statistical models of shape and texture have been widely used for recognition, tracking and synthesis [31, 39, 15, 62], but have tended to only be used with near fronto parallel images. Moghaddam and Pentland [46] describe using view based eigenface models to represent a wide variety of viewpoints. Our work is similar to this, but by including shape variation (rather than the rigid eigen patches) we require ....

M. J. Jones and T. Poggio. Multidimensional morphable models : A framework for representing and matching object classes. International Journal of Computer Vision, 2(29):107-131, 1998.

....2D models. In the following we describe the techniques in more detail and give examples of the model, its ability to estimate pose, to track faces and to synthesize unseen views. 2 Background Statistical models of shape and texture have been widely used for recognition, tracking and synthesis [7, 9, 4, 14], but have tended to only be used with near fronto parallel images. Moghaddam and Pentland [11] describe using viewbased eigenface models to represent a wide variety of viewpoints. Our work is similar to this, but by including shape variation (rather than the rigid eigenpatches) we require ....

M. J. Jones and T. Poggio. Multidimensional morphable models : A framework for representing and matching object classes. International Journal of Computer Vision, 2(29):107-131, 1998.

....to pairs of images. In the following we describe the techniques in more detail and give examples of the models, their ability to synthesize new views and to search unseen images. 2 Background Statistical models of shape and texture have been widely used for recognition, tracking and synthesis [7, 9, 3, 17], but have tended to only be used with near fronto parallel images. Moghaddam and Pentland [12] describe using view based eigenface models to represent a wide variety of viewpoints. Our work is similar to this, but by including shape variation (rather than the rigid eigen patches) we require ....

M. J. Jones and T. Poggio. Multidimensional morphable models : A framework for representing and matching object classes. International Journal of Computer Vision, 2(29):107-131, 1998.

....into two broad categories. Those that have a relatively small number of parameters which control global appearance properties, and those that have a relatively large number of parameters controlling local appearance properties. The former class includes the Morphable Models of Jones and Poggio [5], the face models of Vetter [13] the shape and appearance models of Cootes et.al. 3] the intensity surface models of Nastar et.al. 10] and Turk and Pentland s Eigenfaces [12] amongst others. Here changing any one model parameter can change the whole shape or appearance. The second class, that of ....

....a comprehensive review of work in these elds there are surveys of image registration methods and deformable models in medical image analysis [9] 8] We give here a brief review of more recent relevant work. The closest work to ours are the Multidimensional Morphable Models of Jones and Poggio [5]. These are linear models of appearance which can be matched to a new image using a stochastic optimisation method. The model is built from a set of training images using a boot strapping algorithm. The current model is matched to a new image, optical ow algorithms are used to re ne the t. This ....

[Article contains additional citation context not shown here]

M. J. Jones and T. Poggio. Multidimensional morphable models : A framework for representing and matching object classes. International Journal of Computer Vision, 2(29):107-131, 1998.

.... of these approaches, however, is that they attempt to model the entire image as a single rigid patch making it dicult to model changes in pose and feature location [22, 21, 15] More recently, these technique have been generalized to include schemes for modelling deformations in the image plane [1, 9, 6, 11]. These techniques not only learn a set of allowed variations in the image values, but also a set of allowed variations in pixel location. Nevertheless reliable detection and recognition of images across a wide variety of images is not yet a solved problem. A number of more general statistical ....

M. Jones and T. Poggio, \Multidimensional Morphable Models: A Framework for Representing and Matching Object Classes" in International Journal of Computer Vision, Volume 29, No. 2, August 1998, pp. 107-131.

.... mappings can be represented with radial basis function networks, and can be learned from a set of example images [1] The linear combination of prototypical views has been successfully applied to a number of di#erent problems, like the view invariant recognition of faces [1] for view morphing [7], and for the generation of animation sequences [5] The underlying strategy has been also generalized to three dimensional images [13] These promising results from the application of linear object classes to stationary images has motivated us to investigate if similar ideas can be transferred to ....

M. J. Jones. Multidimensional morphable models: A framework for representing and matching object classes. PhD thesis, Dept. of Computer Science, Cambridge, MA, 1997.

....of stationary images. It was shown that appropriately defined linear combinations of few prototypical images of a three dimensional objects can approximate another view of the object very accurately. This method has been used for the synthesis of new virtual images from example images (e.g. [13, 22, 21]) Meanwhile, morphable models have been used in a broad spectrum of technical applications. Examples are the recognition and synthesis of face images [2, 22] the synthesis of new facial expressions [5] or the simulation of talking faces [8] Morphable models approximate new images by linear ....

....causing transparency effects (e.g. 22] 3 A simple way to define correspondence between image sequences is to calculate usual spatial correspondence between the image pairs that occur at the same points in time. In fact, this strategy has been used in the context of computer animation (e.g. [13, 8, 5]) It leads however not to optimal generalization properties. Natural movement patterns often vary slightly with respect to their timing. It is therefore important to be able to compensate for variations in timing in an efficient way. A simple combination on a frame by frame basis does not capture ....

M. J. Jones. Multidimensional morphable models: A framework for representing and matching object classes. PhD thesis, Massachusetts Institute of Technology, Cambridge, MA, 1997.

....shown in Fig. 10. VI. IMAGE ANALYSIS Learning techniques can also be used for addressing problems of image analysis, requiring the estimation of perceptually meaningful parameters. In the case of faces, for example, these parameters may be associated with mouth shapes and expressions. Following [44], in [45] we constructed a linear morphable model from examples of mouths through the use of optical flow and texture information. Mouths pose difficult problems due to the large visual difference between closed and open mouths. About 2000 images of mouths from one person were collected and 93 ....

....for the texture, and , and for the flow) An example of the linear morphable model at work is shown in Fig. 11. The SVMs for regression are trained on a set generated by estimating the true matching parameters using the computationally intensive stochastic gradient descent algorithm described in [44]. The direct parameter estimation obtained (a) b) c) Fig. 11. LMM at work. a) The model texture has a linear combination of texture prototypes. b) The model flow has a linear combination of the flow prototypes. c) The model image obtained by warping the model texture along the model ....

[Article contains additional citation context not shown here]

M. Jones and T. Poggio, "Multidimensional morphable models: A framework for representing and matching object classes," in Proc. 6th Int. Conf. Computer Vision, Bombay, India, 1998, pp. 683--688.

....Additional support is provided by Eastman Kodak Company, Daimler Benz, Siemens Corporate Research, Inc. and AT T. i Introduction Motivation Amongst the many model based approaches to modeling object classes, the Linear Morphable Model is an important one (Vetter and Poggio [10] Jones and Poggio [4]) It has been been used successfully to model faces, cars and digits. In these applications, the task of matching a novel image to the LMM is achieved through a computationally intensive analysis by synthesis approach. In Jones and Poggio [4] the matching parameters are computed by minimizing ....

....important one (Vetter and Poggio [10] Jones and Poggio [4] It has been been used successfully to model faces, cars and digits. In these applications, the task of matching a novel image to the LMM is achieved through a computationally intensive analysis by synthesis approach. In Jones and Poggio [4], the matching parameters are computed by minimizing the squared error between the novel image and the model image using a stochastic gradient descent algorithm. This technique may take several minutes for matching even a single image. A technique that could compute the matching parameters with ....

[Article contains additional citation context not shown here]

M. Jones and T. Poggio. Multidimensional morphable models: A framework for representing and matching object classes. In Proceedings of the International Conference on Computer Vision, Bombay, India, 1998.

....support is provided by EaR ma Koda k Compa ny, Da mler Benz, Siemens Corpora e Resea rch, Inc.a nd AT T. 1 Introduction Motivation Amongst the many model based approaches to modeling object classes, the Linear Morphable Model is an important one (Vetter and Poggio [10] Jones and Poggio [4]) It has been been used successfully to model faces, cars and digits. In these applications, the task of matching a novel image to the LMM is achieved through a computationally intensive analysis by synthesis approach. In Jones and Poggio [4] the matching parameters are computed by minimizing ....

....important one (Vetter and Poggio [10] Jones and Poggio [4] It has been been used successfully to model faces, cars and digits. In these applications, the task of matching a novel image to the LMM is achieved through a computationally intensive analysis by synthesis approach. In Jones and Poggio [4], the matching parameters are computed by minimizing the squared error between the novel image and the model image using a stochastic gradient descent algorithm. This technique may take several minutes for matching even a single image. A technique that could compute the matching parameters with ....

[Article contains additional citation context not shown here]

M. Jones and T. Poggio. Multidimensional morphable models: A framework for representing and matching object classes. In Proceeding of the International Conference on Computer Vision, Bombay, India, 1998. 9

....not. One criticism of these approaches is that they attempt to model the entire image as a single rigid patch making it difficult to model changes in pose and feature location. More recently these technique have been generalized to include schemes for modelling deformations in the image plane [1, 8, 5, 10]. These techniques not only learn a set of allowed variations in the image values, but also a set of allowed variations in pixel location. Nevertheless, reliable detection and recognition of faces across a wide variety of faces is not yet a solved problem. More recently a number of more general ....

M. Jones and T. Poggio, "Multidimensional Morphable Models: A Framework for Representing and Matching Object Classes" in International Journal of Computer Vision, Volume 29, No. 2, August 1998, pp. 107-131.

....issues are addressed in section 6. 3 Morphable models As described above, our gaze estimation system uses a morphable model to match the eye region of a face and thus extract information about the head orientation and iris position. The morphable model framework has been described in detail in [5, 4]. The next section summarizes the technique. The basic idea is to use example 2D images to build a model which encompasses the 2D shape and texture (gray level values) of all the example images. Figure 2: The experimental setup. 3.1 Summary of morphable models An image I is viewed as a mapping ....

....minimization algorithm could be used. We have chosen to use the stochastic gradient descent algorithm [7] because it is fast and can avoid remaining trapped in local minima. The derivatives required by the stochastic gradient descent algorithm can be calculated straightforwardly and are given in [5]. 3.3 Modeling the eye region Figure 3: Eight of the 100 frontal face prototypes. Because we want our gaze estimation system to be user independent and to handle changes in iris position and head orientation, our model of the eye region must be able to represent all of these sources of ....

Michael J. Jones and Tomaso Poggio. Multidimensional morphable models: A framework for representing and matching object classes. International Journal of Computer Vision, to appear, 1998.

....are not. One criticism of these approaches is that they attempt to model the entire image as a single rigid patch making it di cult to model changes in pose and feature location. More recently these technique have been generalized to include schemes for modelling deformations in the image plane [1, 8, 5, 10]. These techniques not only learn a set of allowed variations in the image values, but also a set of allowed variations in pixel location. Nevertheless reliable detection and recognition of faces across a wide variety of faces is not yet a solved problem. More recently a number of more general ....

M. Jones and T. Poggio, \Multidimensional Morphable Models: A Framework for Representing and Matching Object Classes" in International Journal of Computer Vision, Volume 29, No. 2, August 1998, pp. 107-131.

No context found.

M. Jones and T. Poggio. Multidimensional morphable models: A framework for representing and matching object classes. In Proceedings of the IEEE International Conference on Computer Vision, pages 683--688, 1998.

No context found.

M. J. Jones and T. Poggio. Multidimensional morphable models : A framework for representing and matching object classes. International Journal of Computer Vision, 2(29):107--131, 1998.

No context found.

M. J. Jones and T. Poggio. Multidimensional morphable models : A framework for representing and matching object classes. International Journal of Computer Vision, 2(29):107-131, 1998.

No context found.

M. Jones and T. Poggio. Multidimensional morphable models: A framework for representing and matching object classes. In Proceedings of the IEEE International Conference on Computer Vision, pages 683--688, 1998.

No context found.

M. J. Jones and T. Poggio. Multidimensional morphable models : A framework for representing and matching object classes. International Journal of Computer Vision, 2(29):107--131, 1998.

No context found.

M. J. Jones and T. Poggio. Multidimensional morphable models : A framework for representing and matching object classes. ############# ####### ## ######## ######, 2(29):107-131, 1998.

No context found.

M. Jones and T. Poggio. Multidimensional morphable models: A framework for representing and matching object classes. In Proceedings of the IEEE International Conference on Computer Vision, pages 683--688, 1998.

No context found.

J. Jones, M and T. Poggio, Multidimensional morphable models: a framework for representing and matching object classes, Int. J. Comp. Vision, 29 (1998), pp. 107--131. 46

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