C. Eckes and J. C. Vorbr uggen, Combining Data-Driven and Model-Based Cues for Segmentation of Video Sequences, Proceedings WCNN96, pp. 868--875, San Diego, CA, USA, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....integration of various cues, behaviors or modalities. In the domain of com puter vision the underlying concepts are At 16 Becker et al. tributed Graph Matching [19, 37, 20] integration of multiple scales [38, 39] and integration of mul tiple segmentation cues in a system of interacting spins [34, 10]. On the control side, the most im portant concept is the autonomous refinement of rough preprogrammed schemas [6, 41] by propri oception and autonomously acquired visual information. Regarding software engineering, the consistently object oriented design of the FLAVOR [28] package developed at ....

Christian Eckes and Jan C. Vorbriiggen. Combining data-driven and model-based cues for segmentation of video sequences. In Proceedings WCNN96, pages 868-875. INNS Press & Lawrence Erlbaum Ass., 1996.

....a set of wavelets with 8 directions and 4 frequencies, thus for each vertex we obtain a vector with 32 complex entries. The vector is called a jet. The graphs are generated automatically from the images: first, the object is separated from the background by a segmentation algorithm described in [3] based on the gray level values of the image. Then a grid graph is put on the resulting object segment. 3 Tracking of Object Features For the calculation of the position of an object point in an unfamiliar view from the positions of the point in a small number of neighbouring sample views it is ....

C. Eckes and J. C. Vorbr uggen, Combining Data-Driven and Model-Based Cues for Segmentation of Video Sequences, Proceedings WCNN96, pp. 868--875, San Diego, CA, USA, 1996.

....we use a set of wavelets with 8 directions and 4 frequencies, thus for each vertex we obtain a vector with 32 complex entries, which is called jet. The graphs are generated automatically from the images: first, the object is separated from the background by a segmentation algorithm described in [9], which is based on the gray level values of the image. Then a grid graph is put on the resulting object segment. 2.2 Tracking of Object Features For the calculation of the position of an object point in a novel view from the positions of the point in three sample views it is necessary to have ....

C. Eckes and J. C. Vorbr uggen. Combining Data-Driven and Model-Based Cues for Segmentation of Video Sequences. In Proceedings WCNN96, pages 868--875, 1996.

.... detection [18, 19] Integration has been applied in the spatial domain [4] scale space [20] and time [6] Secondly, there exist some examples in which di erent segmentation cues, such as color and motion, are integrated [21, 22] or where segmentation cues are combined with object recognition [23], but these systems are the exception rather than the rule. These two types of integration are of low and high level, respectively. This work investigates a third, intermediate level of integration where di erent techniques for the same cue are combined. In a rst stage, Gabor wavelets are used ....

C. Eckes and J. C. Vorbruggen. Combining data-driven and model-based cues for segmentation of video sequences. In Proc. World Congress on Neural Networks, pages 868-875, San Diego, CA, September 1996. Intern. Neural Network Soc., Lawrence Erlbaum Assoc. Inc. Mahwah, NJ.

....on integration of various cues, behaviors or modalities. In the domain of computer vision the underlying concepts are At 16 Becker et al. tributed Graph Matching [19, 37, 20] integration of multiple scales [38, 39] and integration of multiple segmentation cues in a system of interacting spins [34, 10]. On the control side, the most important concept is the autonomous refinement of rough preprogrammed schemas [6, 41] by proprioception and autonomously acquired visual information. Regarding software engineering, the consistently object oriented design of the FLAVOR [28] package developed at our ....

Christian Eckes and Jan C. Vorbruggen. Combining data-driven and model-based cues for segmentation of video sequences. In Proceedings WCNN96, pages

.... e.g. Horn Schunck, 1981; Morikawa Harashima, 1993) Secondly, there exist some examples where different segmentation cues, such as color and motion, are integrated, e.g. Poggio et al. 1988; Dubuisson Jain, 1995) or where segmentation cues are combined with object recognition, e.g. (Eckes Vorbr uggen, 1996), but these systems are the exception rather than the rule. These two types of integration are of low and high level respectively. In this work I investigate a third intermediate level of integration which has, to my knowledge, not been investigated before for segmentation from motion. It is the ....

Eckes, C. and Vorbr uggen, J. C. (1996). Combining data-driven and model-based cues for segmentation of video sequences. In Proc. World Congress on Neural Networks, pages 868--875, San Diego, CA. Intern. Neural Network Soc. Lawrence Erlbaum Assoc. Inc. Mahwah, NJ.

....vision cues: colour and edges. The fusion of information between these visual cues is a sequential process. By colour space analysis of images we identify pixels according to their hue values and cluster these pixels into blobs. Our colour segmentation method [3] is data driven and is based on [4]. Labels (spin values) of small image regions (patches) based on Potts spin model encode the segmentation of the image. By comparing neighbouring patches in terms of their hue values we obtain pairwise similarity values of the corresponding spins, which are used in the Potts model framework to map ....

....all spin pairs that are connected. In order to map low similarity to negative interaction and high similarity to positive interaction, the mean similarity h Phii is subtracted from all similarity values to obtain the used interaction W ij ( The similarity function s parameter as described in [4] is set to 110. Fig. 2(b) shows some typical results of our colour segmentation. We obtain an edge description of the scene by employing the Mallat Wavelet Transform [9] as a multiresolution edge detector. We convolute the image I( x) obtained from our colour segmentation process with filters ....

C. Eckes and J. C. Vorbruggen. Combining datadriven and model-based cues for segmentation of video sequences. In Proc. of WCNN96, pages 868-- 875, San Diego, 1996.

....overview of the image preprocessing with Gabor wavelets, and in subsection 2.2.3 we describe how a graph which represents one view is generated from the results of the segmentation and the Gabor transform. 2.2. 1 Segmentation The segmentation method is based on the system of [14] and described in [2]. The segmentation model contains Potts spins with coarse to fine dynamics comparable to real space renormalisation methods often used in theoretical physics. Average intensity is used as the only low level cue, although the system is able to make use of additional cues if they become available. ....

....and high similarity to positive one, we subtract the mean interaction W from all similarity values to obtain the used interaction W ij . We use the Metropolis [9] algorithm at zero temperature with coarse to fine dynamics to let the system relax to a local energy minimum (see figure 2. 3 D and [2] for details) We have used 3 stages and N(1) 1024 as the number of spins in the highest resolution. The number of spins in each resolution is given by N(n) N(1) 2 2(n 1) The segmentation as described may also provide regions, which are regarded as belonging to the object due to their ....

C. Eckes and J. C. Vorbruggen. Combining Data-Driven and Model-Based Cues for Segmentation of Video Sequences. In Proceedings WCNN96, pages 868--875, San Diego, CA, USA, 16--18 September, 1996. INNS Press & Lawrence Erlbaum Ass.

....the scene pointed to by the human hand. To locate object candidates we use two early vision cues: colour and edges. By colour space analysis of images we identify pixels according to their hue values and cluster these pixels into blobs. Our colour segmentation method is data driven and is based on [2]. Labels (spin values) of small image regions (patches) based on Potts spin model encode the segmentation of the image. By comparing neighbouring patches in terms of their hue values we obtain pairwise similarities of the corresponding spins, which are used in the Potts model framework to map the ....

C. Eckes and J. C. Vorbruggen. Combining data-driven and model-based cues for segmentation of video sequences. In Proc. of WCNN96, pages 868--875, San Diego, 1996.

.... Anandan, 1989) and time (e.g. Morikawa Harashima, 1993) Secondly, there exist some examples where different segmentation cues, such as color and motion, are integrated (e.g. Poggio et al. 1988; Dubuisson Jain, 1995) or where segmentation cues are combined with object recognition (e.g. Eckes Vorbr uggen, 1996), but these systems are the exception rather than the rule. These two types of integration are of low and high level, respectively. In this work I investigate a third intermediate level of integration which has, to my knowledge, not been investigated before for segmentation from motion. It is the ....

Eckes, C. and Vorbr uggen, J. C. (1996). Combining data-driven and model-based cues for segmentation of video sequences. In Proc. World Congress on Neural Networks, pages 868--875, San Diego, CA. Int'l Neural Network Soc., Lawrence Erlbaum Assoc. Inc. Mahwah, NJ.

....of each image of the sequence based on graylevel values, which provides the region of the image occupied by the object (see subsection 2.1) In the second part we track landmarks on the object guided by the result of the segmentation. 2. 1 Segmentation The segmentation method is described in [1] and is based on the system of [2] The segmentation model contains Potts spins with coarse to fine dynamics comparable to real space renormalisation methods often used in theoretical physics. Average intensity is used as the only low level cue, although the system A B C D Fig. 1. Overview of ....

....obtain the used interaction W ij . We use the Metropolis [4] algorithm at zero temperature with a coarse tofine dynamics to let the system relax to a local energy minimum (see figure 1D Fig. 2. Tracking of facial landmarks. The frames 1, 10, 20, 30, 40, and 50 of the sequence are shown here. and [1] for the details) We have used 3 stages and N(1) 1024 as the number of spins in the highest resolution. The number of spins in each resolution is given by N(n) N(1) Delta 2 Gamma2(n Gamma1) 2.2 Tracking The tracking procedure we use is described in [5] and based on [6] and [7] Given a ....

C. Eckes, J. C. Vorbruggen, Combining Data-Driven and Model-Based Cues for Segmentation of Video Sequences, Proceedings WCNN96, INNS Press & Lawrence Erlbaum Ass., San Diego, CA, USA, 16--18 September, pp. 868--875, 1996.

No context found.

C. Eckes and J. C. Vorbr uggen, Combining Data-Driven and Model-Based Cues for Segmentation of Video Sequences, Proceedings WCNN96, pp. 868--875, San Diego, CA, USA, 1996.

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