S. Ullman, High-level Vision, 2nd ed. Cambridge, Massachusetts: The MIT Press, 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....and attachment relations that the language module uses. Adding such a system would be a relatively simple extension for the case where it is only producing new concepts bearing these relations. A more ambitious addition would be a recognition system for tokens or instances of types based on [9]. Such a system would work to identify segments of the input with concepts in the conceptual module, so that other visual systems could produce new information for existing concepts. A simpler version of recognition, which might be additionally helpful, would simply associate positions in the ....

Shimon Ullman. High Level Vision. MIT Press, 1996. 55

....Our work differs from feedforward methods in that our method is iterative, and uses features symmetrically to relate the memory to input in both directions. Our approach is also related to work on visual object recognition that combines perceptual organization and top down knowledge (see Ullman[23]) Our model is inspired by Mumford s[11] and Williams and Jacobs [24] use of Markov models of contours for bottom up perceptual completion. Especially relevant to our work is that of Grimes and Mozer[4] Simultaneous with our work ( 8] they use a bigram model to solve anagram problems, in which ....

S. Ullman. High-level Vision, MIT Press, Cambridge, MA. 1996.

....removing irrelevant agents, flags, or landscape features) As we come to a better understanding of the capabilities and limitations of the system under this restriction, it may be relaxed. Our design of the camera planning module is based in part on a visual routines model of intermediate vision [29]. This can be most easily seen when the system generates a visual comparison of two opposing forces: it moves to a point between the forces, at an appropriate distance, and arranges that they are seen with their lowest points touching a common horizontal line. It is straightforward to break this ....

Ullman, S. 1996. High-level vision. Cambridge, MA: The MIT Press.

....Similar problems occur at other occluding boundaries where the pixels may overlap. Although not easily visible in figure 4(d) such overlap occurs at the top of the front box and has been highlighted by insertion of a red line. However, in this case it is easy to use the relative affine depth [2,30] to render only pixels on the nearest surface, essentially [9] by solving for the depth Z at each control point instead of eliminating it as in (3) Finally, we note that, with the present interpolation scheme, we have not reconstructed parts of the image present in only one basis view although, ....

Ullman,S., "High Level Vision", Cambridge, Mass. MIT Press, 1996.

....is a feature vector. Superghost queries seem more natural to people, but less natural to machines since it is not specified in which position each letter appears. Our approach is related to work on visual object recognition that combines perceptual organization and top down knowledge (see Ullman[17]) Our model is inspired by Mumford s[8] and Williams and Jacobs [18] use of Markov models of contours for bottomup perceptual completion. 2 Experiments with Human Subjects In our experiments frequencies of fragments of lengths ranging from two to eight characters were calculated from a weighted ....

S. Ullman. High-level Vision, MIT Press, Cambridge, MA. 1996.

....changes in the scene in real time. 6.1.2 New York Driving Here we present a variant of McCallum s New York driving task, which constitutes an appropriate test domain [35] with roughly 21000 environment states and 2500 observations. The agent s actions and perception are based on visual routines [59, 60] and deictic representation [1, 63] An opportunity to compare the results of learning to human performance on a simulated environment is one of the attractions here. Action Description gaze forward left Look at closest car in lane to the left gaze forward centre Look at closest car in agent s ....

Shimon Ullman. High-level vision. MIT press, 1996.

....of points from the other set, compute the transformation between the two triples, and tally a vote in the corresponding entry of the table. Again the winner entry determines the matching transformation. The complexity of matching a single query set is #### # # # #. In the alignment method [27, 54], for each triplet of points from the query set, and each triplet from the target set, we compute the transformation between them. With each such transformation, all the other points from the target set are transformed. If they match with query points, the transformation receives a vote, and if ....

S. Ullman. High-Level Vision. MIT Press, 1996.

....is constantly occurring in the simple cell arrays E e , which feed into the complex cell arrays C e;y which in turn feed into the F i V2 type arrays. One could even imagine the Q V4 type arrays being activated by the F arrays regardless of the input from M. Using the terminology introduced in Ullman (1996), only those units which are primed by receiving input from M contribute to the summation into S. The object pattern which is excited in the main memory module determines which of the Q arrays will have enhanced activity towards their summation into S. 22 Thus the nal determination of the ....

....The architecture presented here provides a model for visual selection when we know what we are looking for. What if the system is not guided towards a speci c object class. Clearly high level contextual information can restrict the number of relevant object classes to be entertained, see Ullman (1996)) However even in the absence of such information visual selection occurs. One possibility is that there is a small collection of generic representations which are either learned or hardwired, which are useful for generic object background discrimination, and which have dedicated systems for ....

Ullman, S. (1996), High-Level Vision, M.I.T. Press, Cambridge, MA.

....and recognizing 3D objects is a complex and challenging subject in computer vision. A large variety of methods have been proposed for the task of visual object learning and recognition. Among a large variety of recognition methods, three main classes can be distinguished, as summarized by Ullman [1] : Invariant properties and feature spaces, Parts and structural descriptions, and Alignment approach. In this paper, we propose a new scheme for learning and recognizing 3D objects. Our approach lies in parts and structural description approach. It assumes that each object can be decomposed into ....

S. Ullman, High-Level Vision, The MIT Press, 1995

....satisfactory recognition results. 1 Introduction Recognizing 3D objects is a complex and challenging subject in computer vision. A large variety of methods have been proposed for the task of visual object recognition. Recently, for reducing the cost of recognition, recognition by prototype [1][2] has been proposed. A prototype records common information of objects in one object class. In this paper, we give an approach of generating a prototype automatically from learning examples, which is composed of similar, but different objects. In our approach, the prototype is a structural model ....

S. Ullman, High-Level Vision, The MIT Press, 1995

....intensity differences is to gain a measure of photometric invariance. One major difficulty in detecting faces is the 17 Figure 7: Detected edges on a training face under three illuminations. variation in the appearance of faces due to the vagaries of lighting; see for example the discussion in (Ullman 1996). In order to diminish the variation, methods such as those based on neural networks usually require preprocessing (Rowley 1999) for instance subtracting a linear component from the grey level map followed by histogram equalization (Sung Poggio 1998) which can be costly. Instead, the ....

Ullman, S. (1996), High-Level Vision, M.I.T. Press, Cambridge, MA.

.... not only by the shape and the surface properties of the object, but also by its disposition with respect to the observer and the illumination sources, by the optical properties of the intervening medium and the imaging system, and by the presence and location of other objects in the scene (Ullman, 1996). Thus, to detect that two images, which may be taken seconds or years apart, belong, in fact, to the same three dimensional object, the visual system must overcome the influence of a number of factors that affect the way objects look. Possible approaches to separating information on the ....

.... tasks, recognition (knowing a previously seen object as such) appears now to require little more than storing information concerning earlier encounters with the object, as suggested by the success of view based recognition algorithms recently developed in computer vision (Ullman and Basri, 1991; Ullman, 1996). As we shall see, it is surprisingly easy to extend such a memory based strategy to deal with categorization, a task that requires the system to make sense of novel shapes. Thus, familiarity with a relatively small selection of objects can be used as a foundation for processing (i.e. ....

[Article contains additional citation context not shown here]

Ullman, S. (1996). High level vision. MIT Press, Cambridge, MA.

....attention as a benchmark for various learning algorithms. Unlike previous work that emphasizes learning, we approach the problem from a generic perspective that does not involve learning. We point out that the spiral problem is intrinsically connected to the inside outside problem proposed by Ullman (1984, 1996). We propose a solution to both problems based on oscillatory correlation using a time delay network. Our simulation results match human performance, and we interpret human limitations in terms of synchrony and time delays, both biologically plausible. As a special case, our network without ....

....relation is to determine whether a specific pixel lies inside or outside the closed curve. The perception of the inside outside relation is usually effortless for humans. However, the immediate perception is not available for humans when the bounding contour becomes highly convoluted (Ullman 1984, 1996). Ullman (1984) suggested the computation of spatial relation through the use of visual routines. Visual routines result in the conjecture that the inside outside is inherently sequential. As pointed out recently 2 by Ullman (1996) the processes underlying the perception of inside outside ....

[Article contains additional citation context not shown here]

Ullman, S. (1996). High-level vision, MIT Press, Cambridge MA.

....filters are analogous. The principal motivation for using comparisons of intensity differences is to gain a measure of photometric invariance. One major difficulty in detecting faces is the variation in the appearance of faces due to the vagaries of lighting; see for example the discussion in (Ullman 1996). In order to diminish the variation, methods such as those based on neural networks usually require preprocessing (Rowley 1999) for instance subtracting a linear component from the grey level map followed by histogram equalization (Sung Poggio 1998) which can be costly. Instead, the ....

Ullman, S. (1996), High-Level Vision, M.I.T. Press, Cambridge, MA.

....library of source models. In the case of multiple simultaneous sources, where occlusion and masking is very likely to occur, it is important to be able to choose such models based on only limited information. One reasonable approach to solving this problem is to perform hierarchical classification [11]. 6 Musical instrument sounds form a natural hierarchy based on their acoustic propertiesa hierarchy that largely corresponds with the traditional instrument family breakdown. At the highest level, instrument tones are classified as either transient (percussive) or sustained. Sustained sounds ....

S. Ullman, High-level Vision, Cambridge: MIT Press, 1996.

.... extensions of such type of formalism have been proposed in the literature, which include symbolic descriptions of similarities and analogies among entities [10] The algorithms linking together the intermediate and the symbolic representations are generally known as high level vision algorithms [27]. Their role is to identify and classify the entities of the intermediate representation in the terms of the symbolic formalism. In the following, we will concentrate on the intermediate representation block of this schema. In particular, in Sect. 2, we will present the approach proposed in [6, ....

S. Ullman. High-level Vision. MIT Press, Cambridge,MA, 1996.

....or as part of a different longer sequence, thereby providing a mechanism for combining acquired experience to produce novel complex actions. Ordered sequential behavior is important not only in motor control but also in other cognitive domains such as language processing and visual perception (Ullman, 1996). It will be of interest to explore in the future whether common learning mechanisms of sequential behavior could be applied in these domains as well. A Details of the Simulated Plant For the experiments we have used a simulated planar arm with two degrees of freedom (Asada and Slotine, 1986) ....

Ullman, S. (1996). High-Level Vision. Object Recognition and Visual Cognition, chapter 10, pages 317--358. The MIT Press.

No context found.

Shimon Ullman, 1996. High-level vision. Cambridge, MA: MIT Press.

No context found.

S. Ullman, High-level Vision, 2nd ed. Cambridge, Massachusetts: The MIT Press, 1997.

No context found.

S. Ullman (1997) High-level Vision. Object Recognition and Visual Cognition. A Bradford Book, The MIT Press, New York.

No context found.

Shimon Ullman, High-level vision, The MIT Press, Cambridge, MA, 1996.

No context found.

S. Ullman, High-level Vision, MIT Press, Cambridge, MA. 1996.

No context found.

Ullman S.: High-Level Vision, MIT Press #1996#.

No context found.

ULLMAN S. 1996 High-Level Vision, MIT Press.

No context found.

S. Ullman, High level vision, MIT Press, Cambridge, MA, 1996.

First 50 documents  Next 50

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