M. J. Tarr and M. J. Black. A computational and evolutionary perspective on the role of representation in vision. Computer Vision, Graphics, and Image Processing, 60(1):65--73, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....TARGET other moving object RESULT CAMERA Figure 1: Example of an applicability rule for a vision routine and how the information from the approximate world model is used. It is interesting to situate our scheme in the ongoing debate about reconstructionist vs. purposive vision discussed in [13] and in the replies in the same issue. Our proposal falls between the strictly reconstructionist and purely purposive strategies. We are arguing that reconstruction should exist at the approximate level to guide purposive vision routines: by building an approximate model of the scene vision ....

Michael J. Tarr and Michael J. Black. A computational and evolutionary perspective of the role of representation in vision. CVGIP: Image Understanding, 60(1):65--73, July 1994.

.... 200 300 400 100 150 200 250 0 1 2 x 10 3 y (pixel) x (pixel) d) After re locating this feature Figure 8: Evolution of probability maps of position of the focus feature #0 Thus in conclusion, if one of the holy grails of vision science is the development of the general vision system [13, 14], our particular version of the search is for a system architecture in which generality is provided by a Bayesian network (or equivalent probabilistic reasoning system) in which communication is by propagation of revised estimates of probability, and where local control is exercised by the ....

M. J. Tarr and M. J. Black, A computational and evolutionary perspective on the role of representation in vision, CVGIP - image understanding, 60(1), 1994, pp65-73

....systems where new view based routines can be incorporated easily as long as they require similar information from the approximate world model. Reconstructionist vs. Purposive Vision It is interesting to situate our scheme in the ongoing debate about reconstructionist vs. purposive vision (see [21] and the replies in the same issue) Our proposal falls between the strict reconstructionist and purely purposive strategies. We are arguing that reconstruction should exist at the approximate level, guiding the purposive vision routines of the view based level. By making the task routines ....

M. J. Tarr and M. J. Black, "A Computational and Evolutionary Perspective of the Role of Representation in Vision," CVGIP: Image Understanding, vol. 60(1), pp. 65--73, July 1994.

....to check the applicability conditions of view based routines. The approximate world model is not to be used directly as a source of the perceptual information required by specific tasks. It is interesting to situate our scheme in the ongoing debate about reconstructionist vs. purposive vision (see [21] and the replies in the same issue) We are arguing that reconstruction should exist at the approximate level, guiding the purposive vision routines of the view based level of representation. By making the task routines dependent mainly on view based data, we avoid the theoretical and pragmatical ....

M. J. Tarr and M. J. Black, "A Computational and Evolutionary Perspective of the Role of Representation in Vision," CVGIP: Image Understanding, vol. 60(1), pp. 65--73, July 1994.

....trash ooe a AEoor and depositing it in a garbage can. 1 Introduction A longstanding goal in computer vision has been to develop general vision that can support a wide variety of goals, and inform the observer of the relevant ways in which the world does not meet its expectations [ Marr, 1982; Tarr and Black, 1994; Brown, 1994 ] On the other hand, it is advantageous for working systems to make as much use of the task and domain constraints as possible; vision systems that have successfully supported nontrivial tasks have invariably done so (see e.g. Horswill, 1993 ] Dickmanns and Graefe, 1988 ] ....

Michael J. Tarr and Michael J. Black. A computational and evolutionary perspective on the role of representation in vision. CVGIP: Image Understanding, 60:6573, 1994.

....visual information possible Which algorithms are suited at which level of abstraction (pixel, feature, symbol level) S3: paradigms) Interpretation vs. measurements, recognition vs. reconstruction: Much has been written about these different directions in vision (see the dialogue initiated by [26] for a recent example) While the complete 3D reconstruction of complex scenes often is too expensive (and not required) some measurements will be needed in most of the cases. What are the limits of interpretation systems (e.g. remote sensing or medical image interpretation) How can these ....

M.J. Tarr and M.J. Black. A computational and evolutionary perspective on the role of representation in vision. CVGIP Image Understanding, 60(1):65--73, 1994.

....hardly usable by any other task or module. The major problem with this approach is that it is difficult to incorporate high level knowledge into the system, compromising its robustness and adaptability. The reconstructionism vs. purposivism debate is still a central issue in computer vision (see [15] and the replies in the same issue) Our proposal combines both approaches by approxi2 mately reconstructing the world, only to a level such that specific, purposive vision routines can be selected for each task. Contrary to traditional reconstructionism, our approximate world model is not a ....

Michael J. Tarr and Michael J. Black. A computational and evolutionary perspective of the role of representation in vision. CVGIP: Image Understanding, 60(1):65--73, July 1994.

....similar to what we call a skill. Our approach, however, like Horswill s [12] puts more emphasis on taking advantage of constraints that exist in the environment than is suggested in Aloimonos s writings. The article by Aloimonos that explains purposive vision is a reply to Tarr and Black [23], who criticize the purposive approach as being inconsistent with igeneral visionj, which they equate with human vision. Aloimonos points out that even human vision is not entirely igeneral visionj. Yet it is useful for many dioeerent tasks in a wide variety of environments, which is the ultimate ....

Michael J. Tarr and Michael J. Black. A computational and evolutionary perspective on the role of representation in vision. CVGIP: Image Understanding, 60:6573, 1994.

....directly in response to streams of images (e.g. Pomerleau 1993) However, many of these systems do not produce detailed internal symbolic representations of the environment, which we believe are necessary for many types of intelligent behavior. For an interesting discussion on these issues, see Tarr and Black (1994). 7 2. Knowledge Based Image Understanding Simply put, interpreting an image is a matter of establishing correspondences between the image (i.e. the signal) and objects in the knowledge base (the symbols) Thus, it is not sufficient to reconstruct only the geometry of a scene, for example by ....

Tarr, M.J., Black, M.J., A Computational and Evolutionary Perspective on the Role of Representation in Vision, CVGIP(60), No. 1, July 1994, pp. 65-73.

.... On the other side of the issue are those who argue that is too soon to abandon the goals of the recovery paradigm and that, in fact, the recovery approach, with its emphasis on representations, provides the best hope for modeling and understanding general purpose vision in humans and machines [Tarr and Black, 1991] . While recognizing that the purposive paradigm may be appropriate for describing low level, or reflexive, behaviors they level the following criticisms at the approach: ffl the purposive solutions will not scale up to more sophisticated problems, ffl the purposive approach cannot account ....

M. J. Tarr and M. J. Black. A computational and evolutionary perspective on the role of representation in computer vision. Technical Report YALEU/DCS/RR-899, Yale University, October 1991.

No context found.

M. J. Tarr and M. J. Black. A computational and evolutionary perspective on the role of representation in vision. Computer Vision, Graphics, and Image Processing, 60(1):65--73, 1994.

No context found.

M. J. Tarr and M. J. Black. A computational and evolutionary perspective on the role of representation in vision. Computer Vision, Graphics, and Image Processing, 60(1):65--73, 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