E. H. Adelson, J. Y. A. Wang, and S. A. Niyogi. Mid-level vision: new directions in vision and video. In Proc. IEEE International Conference on Image Processing., volume II, pages 21--25, Austin, Texas, 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....major improvements in video coding, a lot of research therefore focused on compression methods that use a higher level of processing, which deals with higher level concepts such us global motions, surfaces, regions, boundaries, textures, etc. These concepts can be referred to as mid level concepts [1] because they are sophisticated enough to produce powerful representations, and yet simple enough to be computed. This higher level of processing can come in many different forms. At one side of the range there are relatively simple methods like applying motion vectors to regions instead of ....

....the transformation are in that work found by block matching, which presents a problem because the transformations do not have just two degrees of freedom but six (affine) or nine (perspective) This increases the search space and therefore the computational load dramatically. Adelson and Wang in [1][21] use the notions of regions and 3 D depth in their work by representing moving scenes with layers. The scene that is shown in the sequence is segmented into layers: one background layer and a number of occluding layers, ordered by depth. As the sequence goes along, the layers are segmented by ....

E. H. Adelson, J. Y. A. Wang, and S. A. Niyogi. Mid-level vision: new directions in vision and video. In Proc. IEEE International Conference on Image Processing., volume II, pages 21--25, Austin, Texas, 1994.

....The video composition in the IERoom can be divided into 3 categories based on the level of exploiting 3D property. One is the 2D composition where multiple image sources are just juxtaposed in layers. The second is the 2. xD composition based on the layered representation of image sequence [9], which is beyond 2D but below 3D. We segment each image into some layers and handle each layer independently. The third is the 3D composition based on full 3D information of a scene, which we focus on in this paper. Suppose we are to merge multiple image sources into one image. If all the image ....

E.Adelson, J.Wang, and S.Niyogi, "Mid-level vision: New directions in vision and video," Proc. IEEE ICIP'94, pp.21-25, Austin, USA, Nov. 1994.

....In fact, under the rigid body assumption, ie that moving bodies do not alter their shape, affine transformations are capable of perfectly modelling the motion resulting from orthographic 3 D to 2 D projection. The idea of using affine transformations for motion estimation purposes is not new [29, 30, 31]. Although it might be possible to determine the full affine transformation directly by search, this has proved computationally expensive [30] Most approaches in use attempts to solve the motion equations directly by adopting some feature based scheme [32, 33] which can involve the solution of as ....

....expensive [30] Most approaches in use attempts to solve the motion equations directly by adopting some feature based scheme [32, 33] which can involve the solution of as many as 15 simultaneous non linear equations. De Castro [34] used Fourier methods to estimate rotation. Adelson et al. [29] use a coarse to fine gradient based motion estimation scheme to determine optical flow. An affine model is then fitted to the found optical flow by means of some least squares fitting procedure. The affine model and estimation scheme used here is based on the frequency domain approach used by Hsu ....

E. H. Adelson, J. Y. A. Wang, and S. A. Niyogi, "Mid-level Vision: New Directions in Vision and Video," in Proceedings IEEE International Conference on Image Processing, vol. 2, (Austin, Texas), pp. 21--25, 1994.

....of the motion. It also serves as a vehicle for the application of standard Kalman filtering techniques [39] to further improve the robustness of the approach. The approach can be viewed as embracing the mid level vision techniques advocated by, amongst others, Adelson, Wang and Niyogi [2, 143, 109], or the idea of a 2 1 2 D sketch as proposed by Marr [96] a sequence is represented at a level which is not the full 3 D object space, but using abstract concepts such as oriented, classified surfaces. The region tracker is shown to successfully track connected surface patches undergoing ....

E. H. Adelson, J. Y. A. Wang, and S. A. Niyogi. Mid-level vision: new directions in vision and video. In Proceedings IEEE International Conference on Image Processing, volume 2, pages 21--25, Austin, Texas, 1994.

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