Jacobs, D., 1989. "Grouping for Recognition." MIT AI Memo 1177.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....to interpret groups of data features such as edges and lines. Previous approaches to selection have focused on the problem of data driven selection by grouping data features such as edges, lines, points, or based on constraints such as parallelism, or collinearity, 19] distance and orientation [18], and regions enclosed by a group of edges [6] The extent to which such grouping methods reduce the search in recognition depends on the reliability of the groups produced (i.e. how many of them really come from a single object) Maintaining the reliability of groups was found to be difficult ....

D.W. Jacobs, "Grouping for recognition," AI Memo. 1177, M.I.T. Artificial Intell. Lab., 1989.

....to interpret groups of data features such as edges and lines. Previous approaches to selection have focused on the problem of data driven selection by grouping data features such as edges, lines, points, or based on constraints such as parallelism, or collinearity, 19] distance and orientation [18], and regions enclosed by a group of edges [6] The extent to which such grouping methods reduce the search in recognition depends on the reliability of the groups produced (i.e. how many of them really come from a single object) Maintaining the reliability of groups was found to be difficult ....

D.W. Jacobs, "Grouping for recognition," AI Memo. 1177, M.I.T. Artificial Intell. Lab., 1989.

....define closed contours (Figure 18 c) they may also define separated curve pieces. It is therefore necessary to find curve pieces which smoothly continue one another and to join them into continuous contours. This problem of grouping curve pieces into curves has been extensively studied (e.g. [45]) Evidently, it does not have a unique solution, and additional constraints have to be used. The most intuitively plausible approach reduces the problem to that of fitting curves of least energy [43, 110] However, this approach can be applied only in simple cases, and cannot handle multiple ....

D.W. Jacobs. Grouping for recognition. AI Memo 1177, MIT, Cambridge, MA, November 1989.

....occludes another. Other methods are therefore needed to tentatively match features. One possibility would be 2D matching that incorporated geometric constraints and perhaps included a low level perceptual grouping strategy that can identify and group features originating from a common rigid object [9, 12]. Reliably detecting corners was also difficult. A coupled feature detection and tracking mechanism, e.g. 19, 20] should be investigated further. The corners were tracked using simple linear Kalman filters. Tuning the various parameters, e.g. process and measurement noise, was straightforward ....

D. W. Jacobs. Grouping for recognition. AI Memo 1177, MIT, 1989.

....images of the object. We find corner features by making straight line approximations to the edges[21] and lo cating a corner where nearby lines have a stable intersection point, when extended. We 14 then form by hand groups of three to five points that are formed by a set of convex lines (see [13] and [12] for discussion of the value of convex groups) The convexity of the group orders the points, although it does not tell us which point comes first. So a different group is formed for each possible starting point. To allow for the effects of occlusion, we also form groups in which any one ....

....on a generaJ approach to recognition that links grouping and indexing. In this approach, a bottom up process forms salient groups in the images, and then matches 19 them to only those model groups that could have produced them. This approach was first used by [19] and has since been taken in: [13], 22] 12] 27] 6] and [7] A number of recent indexing systems for recognizing planax objects from arbitrary 3D views have been based on invariant functions of the image [28] 17] 18] 9] 10] 25] For example, as we have noted, the affine coordinates of a planar model are invariant ....

Jacobs, D., 1989. "Grouping for Recognition." MIT AI Memo 1177.

....parallel lines, and that lines coming from different objects have random relative orientations. Because these types of clues are probabilistic, one expects to achieve better performance by combining many clues together, and this seems to be the experience of many researchers (Lowe[34] Jacobs[25], Shashua and Ullman[46] Denasi et al. 13] Sarkar and Boyer[43] Mohan and Nevatia[38] Williams[49] and Nitzburg and Mumford[39] Lowe[34] also stressed the importance of grouping to object recognition. He pointed out that by grouping together features that are likely to have been all ....

....order one s search for that object. If one looks first at these groups, one can achieve significant speedups in a recognition system, compared to performing a random search. Others have also explicitly used grouping, or the formation of more complex features, to speed up recognition systems (Jacobs[24, 25], Califano and Mohan[8] Clemens[10] SyedaMahmood [47] Burns and Riseman[7] Huttenlocher and Wayner[23] and Wayner[48] The interaction between grouping and the computational complexity of recognition is treated more theoretically by Grimson[17] and Clemens and Jacobs[11] Most past work on ....

[Article contains additional citation context not shown here]

Jacobs, D., 1989, "Grouping for Recognition," MIT AI Memo 1177.

No context found.

--364. Jacobs, D. W. (1989). Grouping for recognition (A.I. Memo No.

No context found.

Vision 8(1): 7--29. Jacobs, D.W. 1989. Grouping for Recognition . A.I. Memo No. 1177, A.I. Lab., Mass. Inst. Technol.

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