J. Weng, N. Ahuja, and T. S. Huang, Learning recognition and segmentation of 3-d objects from 2-d images, Proceedings of the International Conference on Computer Vision, ICCV 93, pages 121-128, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....representation is classi ed using a multi layer perceptron. The results received were favorable, but the database was quite simple: the pictures are manually aligned and there is no lighting variation, tilting, or rotation, and there were only 20 people in the database. Weng, Ahuja and Huang [5] make use of an hierarchical neural network which is grown automatically and not trained on the traditional gradient descent method. They reported good results, but once again the database was very small. Samaria and Harter [6] made use of a Hidden Markov Model (HMM) based approach for the ....

J. Weng, N. Ahuja, and T. S. Huang, Learning recognition and segmentation of 3-d objects from 2-d images, Proceedings of the International Conference on Computer Vision, ICCV 93, pages 121-128, 1993.

....lines, and face matching 34 CHAPTER 2. PREVIOUS WORK is performed using correlation on these binary images. Connectionist approaches to face recognition also use pictorial representations for faces (Kohonen[80] Fleming and Cottrell [53] Edelman, Reisfeld, and Yeshurun[49] Weng, Ahuja, and Huang[135], Fuchs and Haken[55] 56] Stonham[124] and Midorikawa[94] Since the networks used in connectionist approaches are just classifiers, these approaches are similar to the ones described above. In multilayer networks of simple summating nodes, inputs such as grey level images are applied at an ....

....connectionist approaches to face recognition, the two most important issues are input representation at the input layer and the overall network architecture. As previously mentioned, the input representations are pixel based, with [80] 53] 55] and [94] using the original grey level images. [135] uses directional edge maps, 124] uses a thresholded binary image, and [49] uses Gaussian units applied to the grey level image. A variety of network architectures have been used. A vanilla multilayer network trained by backprop, probably the most standard approach, has been explored by [53] and ....

[Article contains additional citation context not shown here]

John J. Weng, N. Ahuja, and T.S. Huang. Learning recognition and segmentation of 3-D objects from 2-D images. In Proceedings of the International Conference on Computer Vision, pages 121--128, Berlin, May 1993.

....each new primitive describes a commonly observed configuration of two distinguished points. The induction process is repeated with these new primitives to generate higher level primitives describing groups of four, eight, and more distinguished points. Weng et al. s Cresceptron system [31] is analogous in that it induces a hierarchy of primitives within a pre programmed framework. Since these primitives are essentially templates, invariance to translation, rotation, and scaling in the image must be provided by prior segmentation or by an attentional mechanism. We would expect both ....

J. J. Weng, N. Ahuja, and T. S. Huang. Learning recognition and segmentation of 3-D objects from 2-D images. In Proc. Int. Conf. Comput. Vision, pages 121--128, 1993.

.... Pictorial approaches, representing faces by using filtered images of model faces, include template based systems ( 2] 6] 13] 7] and [5] systems using principal components analysis to derive a pictorial face space ( 15] 20] 1] 9] and connectionist approaches ( 16] 11] 10] [21], and [12] 18] explores an interesting hybrid representation that combines the geometrical and pictorial approaches, representing faces as elastic graphs of local textural features. The wide variation in face appearance under changes in pose, lighting, and expression makes face recognition a ....

John J. Weng, N. Ahuja, and T.S. Huang. Learning recognition and segmentation of 3-D objects from 2-D images. In Proceedings of the International Conference on Computer Vision, pages 121--128, Berlin, May 1993.

....role in the development of such a capability [vision] in humans. The evidence for learning in vision includes even low level vision. For instance, a common visual experience, overhead light source, is learned and used to perceive shape from shading [reference to Ramachandran] 1) 1 [38]. We shall return to this work in Section 6, but I should note here that the learning model proposed in [38] is one that indeed extends the architecture of the basic neural net model by processes that add structure to the net, so it incorporates an important aspect of development into its ....

....vision includes even low level vision. For instance, a common visual experience, overhead light source, is learned and used to perceive shape from shading [reference to Ramachandran] 1) 1 [38] We shall return to this work in Section 6, but I should note here that the learning model proposed in [38] is one that indeed extends the architecture of the basic neural net model by processes that add structure to the net, so it incorporates an important aspect of development into its interpretation of learning . It seems to me that more vision researchers than not believe, or assume, or are forced ....

[Article contains additional citation context not shown here]

T. S. Huang J. J. Weng, N. Ahuja. Learning recognition and segmentation of 3-D objects from 2-D images. In International Journal of Computer Vision, pages 121--128, 1993.

....is quite simple: the pictures are manually aligned and there is no lighting variation, rotation, or tilting. There are 20 people in the database. A hierarchical neural network which is grown automatically and not trained with gradient descent was used for face recognition by Weng and Huang [44]. They report good results for discrimination of ten distinctive subjects. 3.6 The ORL Database In [39] a HMM based approach is used for classification of the ORL database images. The best model resulted in a 13 error rate. Samaria also performed extensive tests using the popular eigenfaces ....

J. Weng, N. Ahuja, and T.S. Huang. Learning recognition and segmentation of 3-d objects from 2-d images. In Proceedings of the International Conference on Computer Vision, ICCV 93, pages 121--128, 1993.

.... Singular Value Decomposition (Cheng, et al. 11] and Hong[19] and vector quantization (Ramsay, et al. 31] Connectionist approaches to face recognition also use pictorial representations for faces (Kohonen[24] Fleming and Cottrell [16] Edelman, Reisfeld, and Yeshurun[15] Weng, Ahuja, and Huang[40], Fuchs and Haken[17] Stonham[36] Since the networks used in connectionist approaches are just classifiers, these approaches are similar to the ones described above. Different pixel based representations have been used, with [24] 16] 17] using the original grey level images. 40] uses ....

....and Huang[40] Fuchs and Haken[17] Stonham[36] Since the networks used in connectionist approaches are just classifiers, these approaches are similar to the ones described above. Different pixel based representations have been used, with [24] 16] 17] using the original grey level images. [40] uses directional edge maps, 36] uses a thresholded binary image, and [15] uses Gaussian units applied to the grey level image. Hybrid representations that combine the geometrical and pictorial approaches have been explored, such as Cannon et al. 9] whose feature vector face representation ....

John J. Weng, N. Ahuja, and T.S. Huang. Learning recognition and segmentation of 3-D objects from 2D images. In Proceedings of the International Conference on Computer Vision, pages 121--128, Berlin, May 1993.

....the approximate region of interest can be identified. This condition is believed to be true for a large class of medical images. Thus, we can avoid a much higher computational cost with methods that are designed for images of virtually unconstrained real world scenes (see, e.g. the Cresceptron [34] and SHOSLIF [33] systems) The general approaches presented in this paper, including multi stage learning, interstage feedback, and system failure detection using knowledge based probability measure, may be useful for other medical image analysis problems. Acknowledgments The authors would like ....

J. Weng, N. Ahuja, and T. S. Huang. Learning recognition and segmentation of 3-d objects from 2-d images. In Proc. 4th International Conf. Computer Vision, pages 121--128, Berlin, Germany, May 1993. 30

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