Williams, L. R. and Hanson, A. R. (1994). Perceptual completion of occluded surfaces. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pages 104112. IEEE Comput. Soc. Press.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....is taken among all curves C joining p and q with tangents # p and # q , respectively, k denotes the curvature of C, ds its arc length, and #, # are positive constants. Let us mention that Euler s elastica has been frequently used in computer vision ( 41] 50] 64] 69] 70] 71] 75] 76] [74]) and a beautiful account on it can be found in [58] In an important contribution to the question Masnou and Morel [52, 54, 53] proposed a variational formulation for the recovery of the missing parts of a grey level two dimensional image and they referred to this interpolation process as ....

L.R. Williams and A.R. Hanson, Perceptual completion of occluded surfaces, Proceedings of the IEEE Computer Vision and Pattern Recognition, Seattle, WA, pp. 104--112, 1994.

....very likely that they can not di erentiate between the cases in Figs 1A and 1B. Models that were constructed by a surface reconstruction scheme based on the judgment of occlusions, on the other hand, showed more robust properties re ecting the di erences between the variations of the gure [17] [18]. Our model, which is based on the analysis of junction properties to construct a depth map, will be closer to the surface completion models (see also the hybrid model [19] We, however, corroborate the rather intuitive point of departure of surface completion models by arguments obtained from ....

Williams, L., Hanson, A.: Perceptual Completion of Occluded Surfaces. Comput. Vis. Image. Understand. 64 (1996) 1-20

....energy have been introduced a constraint, to be sure, but one whose motivation is less clear than sso. In addition, these global procedures are typically topologically constrained to ensure that the elastica complete descriptions of surfaces that make physical sense as smooth manifold solids [57] or paper cutouts [36] Closure The paper cutout constraint is equivalent to the requirement that the bounding contour of every object is a closed curve. That figural closure is connected to the grouping problem is revealed in Fig. 6, where the three bow ties to the right are usually more salient ....

L. R. Williams. Perceptual Completion of Occluded Surfaces. PhD thesis, University of Massechusettes at Amherst, 1994.

....stronger global constraint is required. Motivated by growing evidence that the human visual system exploits contour closure for the purposes of perceptual grouping [6, 7, 14, 15, 25] we present an algorithm for computing highly closed bounding contours from images. Unlike previous algorithms [11, 18, 26], no restrictions are placed on the type of structure bounded or its shape. Contours are represented locally by tangent vectors, augmented by image intensity estimates. A Bayesian model is developed for the likelihood that two tangent vectors form contiguous components of the same contour. ....

....extended chains. However, no attempt is made to compute closed chains; a necessary condition for computing global 2 D shape properties and for segmenting structures from an image. A separate branch of research investigates the grouping of occlusion edges into complete contours, ordered in depth [18,26]. While interesting from a theoretical point of view, a large fraction of the edges in real images are not occlusion edges, and a recent study [5] suggests that it is not possible to locally distinguish occlusion edges from other types of edges. It is our view that algorithms for grouping contours ....

[Article contains additional citation context not shown here]

L.R. Williams. Perceptual completion of occluded surfaces. PhD thesis, University of Massachusetts, Amherst, Mass., February 1994.

....images is one of the objectives of perceptual grouping. By perceptual grouping, we refer to the (bottom up) process of grouping together structures in the image that are likely to belong to a single object. Other tasks in perceptual grouping are image segmentation and gap completion. For instance, [13, 15, 17, 18, 19, 20, 22, 23, 31, 32] extract contours from the image according to certain optimization criteria, 29, 24, 3, 14, 5] compute optimal curves for filling in gaps, and [4, 30, 8, 10, 12, 21, 33] identify occluded and subjective contours. In this paper we provide an analysis of Shashua and Ullman s method. We examine ....

....becomes at least cubic in the size of the image. As mentioned above, we also consider the possibility of using the Saliency Network for grouping. We note that, in contrast to other existing methods for grouping that search over the exponentially large space of all possible image curves (e.g. [13, 15, 22, 32]) the Saliency Network recovers the most salient curve in time complexity that is polynomial in the size of the image. However, the network must take a single choice at every junction, and curves lying near a salient curve tend to merge into the salient curve because they can benefit from its ....

Williams, L. R., and A. R. Hanson, "Perceptual Completion of Occluded Surfaces," Proc. IEEE Conf. Computer Vision Pat. Rec., Seattle, WA, June 1994. Also to appear in Computer Vision and Image Understanding.

....a scene model consisting of a stick, a triangular card, and a quadrilateral card. The triangular card is in front of both the stick and the quadrilateral. For objects which do not intersect in the image, such as the stick and the quadrilateral, the depth relation is taken to be undefined (see [15] for a more general 2D layered scene model) Qualitative priors can be derived for such scene models consisting of multiple objects. For the current paper we take the shape and position of any object to be independent of the shape and position of other objects. For example, the scene model ....

....ordering on interpretations for simple card world and blocks world scenes. Moreover the analysis motivated the choice of effective search heuristics. The same style of analysis can be applied to other domains, such as model based object recognition [3, 5, 7, 10] curve and surface grouping [4, 14, 15], and simple motion interpretation [8] These are important areas for further study. An open question concerns how our approach based on qualitative probabilities performs compared to quantitative approaches for scene interpretation (e.g. 1, 3, 7, 9] and further if we can exploit quantitative ....

L. Williams and A. Hanson. Perceptual completion of occluded surfaces. CVIU, 64(1):1--20, 1996.

.... , Kumaran, Geiger and Gurvits[10] The discussion on visual organization dates back at least to the labeling scheme by Huffman [7] These approaches are too weak to constraint in the sense that they allow for far to many possible organizations (all of them equally likely) Williams and Hanson [16] and Williams and Rubin [17] propose a linear programming method based on local information at junctions, that is partially considered here. However, they do not take into account neither region information nor local properties of the reconstructed illusory contours. Shashua and Ullman [14] do ....

L. R.Williams and A. Hanson. Perceptual completion of occluded surfaces. CVPR, 1994.

.... constants between interacting entities and then formulate perceptual organization as a combinatorial optimization problem which can be solved by using techniques borrowed from statistical mechanics [16, 12] A combinatorial formulation by means of integer programming has also been proposed [33, 34]. 3 Description of the algorithm As Figs. 2 and 3 illustrate, a typical contour graph contains several vertices with high degree. As a consequence, the number of paths in the graph is exponentially large and it is not possible to test them all to detect the salient ones. Dynamic programming ....

L.R. Williams and A.R. Hanson. Perceptual completion of occluded surfaces. In IEEE Computer Vision and Pattern Recognition, pages 104--112, 1994.

....images is one of the objectives of perceptual grouping. By perceptual grouping, we refer to the (bottom up) process of grouping together structures in the image that are likely to belong to a single object. Other tasks in perceptual grouping are image segmentation and gap completion. For instance, [9, 11, 12, 13, 14, 15, 17, 18, 25, 26] extract contours from the image according to certain optimization criteria, 23, 19, 1, 10, 3] compute optimal curves for filling in gaps, and [2, 24, 5, 6, 8, 16, 27] identify occluded and subjective contours. In this paper we provide an analysis of Shashua and Ullman s method. We examine both ....

....the complexity of the network becomes cubic in the size of the image. Finally, we consider the possibility of using the Saliency Network for grouping. We note that, in contrast to other existing methods for grouping that search over the exponentially large space of all possible image curves (e.g. [9, 11, 17, 26]) the Saliency Network recovers the most salient curve in time complexity that is polynomial in the size of the image. However, the network must take a single choice at every junction, and as a consequence has problems with identifying salient curves other than the most salient one. The paper ....

Williams, L. R., and A. R. Hanson, "Perceptual Completion of Occluded Surfaces," Proc. IEEE Conf. Computer Vision Pat. Rec., Seattle, WA, June 1994. Also to appear in Computer Vision and Image Understanding.

.... constants between interacting entities and then formulate perceptual organization as a combinatorial optimization problem which can be solved by using techniques borrowed from statistical mechanics [14, 10] A combinatorial formulation by means of integer programming has also been proposed [30, 31]. 1.2 Contributions of this paper Ideally, a contour estimation algorithm should compute a set of contour descriptors which contains an accurate approximation to every contour in the scene. Also, it would be desirable to have probability estimates associated to these contour descriptors. A ....

L.R. Williams and A.R. Hanson. Perceptual completion of occluded surfaces. In IEEE Computer Vision and Pattern Recognition, pages 104--112, 1994.

....from image contours (or a line drawing) by means of the three stage process depicted in Figure 2(a) The first stage is figural completion, which is the process of inferring a set of interpolating curves, or completions, satisfying the Huffman labeling scheme. The author s recent Ph.D. thesis[13] describes a system which solves figural completion problems in the anterior scene domain (i.e. Figure 1(a) and (b) An anterior scene is a set of smooth surfaces with boundary embedded in three dimensional space such that the surface normals everywhere have a positive component in the viewing ....

....labeling scheme for smooth objects. Single arrows represent boundaries and double arrows represent occluding contours. Numbers are depth indices and direction of arrows indicate sign of occlusion. a) and (b) Boundary crossing junctions are sufficient to represent the domain of anterior scenes[13]. An anterior scene is a set of smooth surfaces with boundary embedded in three dimensional space such that the surface normals everywhere have a positive component in the viewing direction. c) through (f) Occluding contour crossing junctions and cusp junctions together define the domain of ....

Williams, L.R., Perceptual Completion of Occluded Surfaces, Ph.D. dissertation, Dept. of Computer Science, University of Massachusetts, Amherst, Mass., 1994.

....considered necessary in the sense that the image of the boundary of any anterior scene satisfies the constraints. But does a set of closed contours satisfying the labeling scheme always define an anterior scene Is the labeling scheme necessary and sufficient In his recent Ph.D. thesis, Williams[12] shows that this is in fact the case. Specifically, he shows that every knot diagram satisfying the labeling scheme illustrated in Figure 1 represents a generic view of an anterior scene. The proof is based upon the existence of a procedure for building a combinatorial model (i.e. a paneling[2] ....

.... a unique surface organization, then additional preference criteria must be employed (see Rock[9] Unfortunately, it is difficult to design an objective function which incorporates these criteria without introducing a bias for (or against) unit organizations with larger numbers of completions (see [12]) Solving this problem is a subject for future work. 2.2 Experimental system To simplify implementation of the experimental system, off the shelf components were used wherever possible. For example, the straight line grouping algorithm developed by Boldt[1] was used to generate the input set ....

[Article contains additional citation context not shown here]

Williams, L.R., Perceptual Completion of Occluded Surfaces, Ph.D. Dissertation, Dept. of Computer Science, Univ. of Massachusetts at Amherst, Amherst, Mass., 1994.

....boundaries which are nominally visible (i.e. no occluding surface is presumed to be hiding them) Illusory contours are therefore modal completions. Amodal completions are nominally occluded (i.e. an intervening surface is presumed to be hiding them) Kellman and Shipley[15] and others (e.g. [22, 33]) hypothesize that modal and amodal completions are computed by the same underlying process. According to this view, whether or not a completion is perceived as modal or amodal is determined by a later process devoted to organizing surfaces. We are not concerned here with all of the factors ....

....surfaces. We are not concerned here with all of the factors which play a role in determining whether or not a completion will be perceived as modal or amodal. Some of these factors are beyond the scope of the current paper (they concern the topological validity of the surface organization[33]) Rock[26] emphasizes one factor, termed stimulus conformity, which is relevant to our current discussion. Rock s notion of stimulus conformity can be used to explain why illusory contours (unlike amodal completions) cannot cross large brightness discontinuities. This is based on the following ....

Williams, L.R., Perceptual Completion of Occluded Surfaces, Ph.D. Dissertation, Dept. of Computer Science, Univ. of Massachusetts at Amherst, Amherst, Mass., 1994.

No context found.

Williams94a Williams, L.R., Perceptual Completion of Occluded Surfaces, Ph.D. Dissertation, Dept. of Computer Science, Univ. of Massachusetts at Amherst, Amherst, Mass., 1994.

No context found.

Williams, L. R. and Hanson, A. R. (1994). Perceptual completion of occluded surfaces. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pages 104112. IEEE Comput. Soc. Press.

No context found.

L. R. Williams, Perceptual Completion of Occluded Surfaces, Ph.D. thesis, University of Massachusetts at Amherst, 1994.

No context found.

L. Williams and A. Hanson, Perceptual completion of occluded surfaces, Comput. Vision Image Understanding 64, 1996, 1--20.

No context found.

L. R. Williams and A. R. Hanson, "Perceptual Completion of Occluded Surfaces," Computer Vision and Image Understanding, vol. 64, no. 1, pp. 1-20, 1996.

No context found.

L. R. Williams, and A. R. Hanson, "Perceptual Completion of Occluded Surfaces," Proc. of IEEE Computer Vision and Pattern Recognition, Seattle, WA., 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