R. Bolles, H. Baker, and M. Hannah. The JISCT Stereo Evaluation. In DIUW, pages 263--274, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....recent study by Mitiche and Bouthemy [78] reviews a large number of methods for image flow computation and isolates central problems, but does not provide any experimental results. In stereo correspondence, two previous comparative papers have focused on the performance of sparse feature matchers [54, 19]. Two recent papers [111, 80] have developed new criteria for evaluating the performance of dense stereo matchers for image based rendering and tele presence applications. Our work is a continuation of the investigations begun by Szeliski and Zabih [116] which compared the performance of several ....

....cost over square windows with constant disparity; 3. disparities are computed by selecting the minimal (winning) aggregated value at each pixel. Some local algorithms, however, combine steps 1 and 2 and use a matching cost that is based on a support region, e.g. normalized cross correlation [51, 19] and the rank transform [129] This can also be viewed as a preprocessing step; see Section 3.1. On the other hand, global algorithms make explicit smoothness assumptions and then solve an optimization problem. Such algorithms typically do not perform an aggregation step, but rather seek a ....

[Article contains additional citation context not shown here]

R. C. Bolles, H. H. Baker, and M. J. Hannah. The JISCT stereo evaluation. In DARPA Image Understanding Workshop, pages 263--274, 1993.

....of the independent method. The semi manual ground truth must be obtained from independent images, not the data set itself. Nine images were used for the Lab Scene [20] for instance. There have been several medium to large scale e#orts to evaluate stereo matching algorithms in a systematic way [11, 8, 1, 25, 23]. In [11] ten di#erent stereo algorithms were re implemented and evaluated. Their comparison was based on the number of correctly matched pixels and thus only measures the overall quality of the matching. The choice of the test images is not focused on any particular application nor motivated ....

....was focused on cartographic feature extraction application. This strongly influenced the test data selection. Only two algorithms based on di#erent matching techniques (area based and feature based approaches) were tested. For a long time unsurpassed evaluation study has been the JISCT e#ort [1]. It is also application oriented but methodologically very advanced. On a large set of di#erent stereo images (44) from real complex scenes, various approaches from di#erent groups (INRIA, SRI, Teleos) were statistically evaluated based on three types of errors: false negatives, false ....

Robert C. Bolles, H. Harlyn Baker, and Marsha Jo Hannah. The JISCT stereo evaluation. In Proc. DARPA Image Understanding Workshop, pages 263--274, 1993. 3

....of the independent method. The semi manual ground truth must be obtained from independent images, not the data set itself. Nine images were used for the Lab Scene [57] for instance. There have been several medium to large scale e#orts to evaluate stereo matching algorithms in a systematic way [37, 29, 5, 66, 61]. 19 In [37] ten di#erent stereo algorithms were re implemented and evaluated. Their comparison was based on the number of correctly matched pixels and thus only measures the overall quality of the matching. The choice of the test images is not focused on any particular application nor motivated ....

....was focused on cartographic feature extraction application. This strongly influenced the test data selection. Only two algorithms based on di#erent matching techniques (area based and feature based approaches) were tested. For a long time unsurpassed evaluation study has been the JISCT e#ort [5]. It is also application oriented but methodologically very advanced. On a large set of di#erent stereo images (44) from real complex scenes, various approaches from di#erent groups (INRIA, SRI, Teleos) were statistically evaluated based on three types of errors: false negatives, false ....

Robert C. Bolles, H. Harlyn Baker, and Marsha Jo Hannah. The JISCT stereo evaluation. In Proc. DARPA Image Understanding Workshop, pages 263--274, 1993.

....by a surface fitting step, e.g. using triangulation, splines, or seed and grow methods. Our work is motivated by a similar study of optical flow algorithms by Barron et al. 5] In stereo correspondence, two previous comparative papers have focused on the performance of sparse feature matchers [35, 15]. Two recent papers [69, 49] have developed new criteria for evaluating the performance of dense stereo matchers for imagebased rendering and tele presence applications. This work is a continuation of the investigations begun by Szeliski and Zabih [72] which compared the performance of several ....

....[39] More recently, robust measures, including truncated quadratics and contaminated Gaussians have been proposed [11, 12, 63] These measures are useful because they limit the influence of mismatches during aggregation. Other traditional matching costs include normalized cross correlation [34, 60, 15], which behaves similar to sum of squared differences (SSD) and binary matching costs (i.e. match no match) 46] based on binary features such as edges [33] or the sign of the Laplacian [51] Binary matching costs are not commonly used in dense stereo methods, however. Some costs are ....

R. C. Bolles, H. H. Baker, and M. J. Hannah. The JISCT stereo evaluation. In IUW, pp. 263-274, 1993.

....be tackled by any algorithm unless a global model or regularization assumptions are made. The error classification above is data driven and given the input images for a stereo algorithm, image areas can be pre classified and results can be evaluated in separate areas as in the JISCT experiment [3] and in [12] All of the above cases appear in our data and reflect in all three error metrics we use. We perform a boolean classification of image areas regarding occlusions and a continuous classification of image areas regarding intensity variation based on an image gradient threshold. 4 Depth ....

R. Bolles, H. Baker, and M. Hannah. The jisct stereo evaluation. In DARPA Image Understanding Workshop, pages 263--274, 1993.

....3.3. Real World Natural images The last tests use real world data. We consider the performance of the occlusion detection algorithms on two standard stereo vision data sets: the pentagon stereo pair from Carnegie Mellon s VASC Image Database and the birches stereo pair from the JISCT test set [5](see Fig. 5) For the Pentagon case, all of the algorithms perform reasonably in signaling the major occlusions arising from the top edge of the building against the ground. Further, under PPP, the region detection algorithms, LRC and ORD, provide markedly less spurious signal than the border ....

R. Bolles, H. Baker, and M. Hannah. The JISCT stereo evaluation. In Proc. DARPA IUW, pages 263--274, 1993.

....to be derived from the image signals used for stereo matching: the more discriminative the image features, the more con#dent the reconstructed terrain point. Generally, the con#dence measures re#ect not only the image and terrain features but also the features of stereo matching. For example, in #Bolles et al. 1993# two con#dence measures for a DPM obtained by a correlation based local optimization are considered. The #rst measure is based on a di#erence p#x; y# # p 0 #x; y# between the best match disparity p#x; y# in the DPM and the second best match disparity p 0 #x; y# for the same planar position ....

R. C. Bolles, H. H. Baker, and M. J. Hannah, The JISCT stereo evaluation. In Proceedings of the DAPRA Image Understanding Workshop, April 18-21, 1993, Washington, D.C., pages 263-274. San Mateo, 1993. Morgan Kaufmann.

....of both methods. The amount of information drawn from the passive method is currently rather small. It remains to be seen how the evaluation of additional processing results from passive methods can be incorporated to improve the results further. To this end some of the methods as described in [8, 9, 10, 11] are currently being implemented and will be tested on scenes similar to that shown in Fig. 3. Moreover, the rules described in section 4.2.3 will be refined and heuristics for the selection of thresholds will be developed. ....

Bolles, R.C., Baker, H.H., and Hannah, M.J., "The JISCT stereo evaluation", in Proc. 1993 Image Understanding Workshop, DARPA, University of Maryland Research Projects, pp. 263-274, 1993.

....small region would be useful as it would reduce the computational cost of stereo matching. In general it is not possible to match all regions of an image pair using passive stereo. However, there are large areas of many image pairs whose depth that can be reliably computed through passive stereo [4]. There are points which can be readily matched and others whose depth can only be estimated by modeling the surface and lighting conditions and interpolating. For a variety of reasons it is not possible to match a single pixel in one image with its counterpart in another image. Therefore some ....

....a rotating polarizing filter. Once the specularities have been removed, stereo matching can proceed. But this requires extensive computation and special hardware. In the case of occlusion, matching is impossible, because the corresponding pair simply isn t there. Some stereo matching algorithms [4] search for an onto mapping between the images, rejecting any matches that do not work from left to right and right to left. This is helpful in eliminating unmatchable points from further consideration. Chung describes an edge based method which classifies different types of occlusions based on ....

Robert C. Bolles, Harlyn H. Baker, and Marsha Jo Hannah. The jisct stereo evaluation. In Proceedings: Image Understanding Workshop, April 1993. 59

....a man and kids. The largest occlusion region in this image is 93 pixels wide, or 13 percent of the image. test imagery. In our lab, common images like Figure 1 contain disparity shifts and occlusion regions over eighty pixels wide. 1 Popular stereo test images, however, like the JISCT test set[9], the pentagon image, the white house image, and the Renault part image have maximum occlusion disparity shifts on the order of 20 pixels wide. Regardless of camera configuration, images of the everyday world will have substantially larger occlusion regions than aerial or terrain data. Even ....

....GCPs. Of course, we are limited to how small the occlusion cost can be. If it is smaller than the typical value of correct matches (non zero due to noise) 6 then the algorithm proposes additional occlusion regions such as in path d of Figure 7. For real stereo images (such as the JISCT test set [9]) the typical DSI value for incorrectly matched pixels is significantly greater than that of correctly matched ones and performance of the algorithm is not particularly sensitive to the occlusion cost. Also, we note that while we have attempted to remove smoothing influences entirely, there are ....

[Article contains additional citation context not shown here]

R. Bolles, H. Baker, and M. Hannah. The JISCT stereo evaluation. In Proc. Image Understanding Workshop, pages 263--274, 1993.

....at every pixel. The work described here is simpler but restricted to the case where a global, spatially invariant, non linear, monotonically increasing relationship exists between the intensities of the two images. This paper first examines the images contained in the SRI JISCT stereo database [2]. Section (2) reveals that the constant image brightness assumption is often false. Moreover, neither a DC bias nor a linear model adequately represents the observed relationships. A comprehensive 2 physical model of the observed deviations is difficult to develop. In fact, it is unlikely that a ....

R. C. Bolles, H. H. Baker, and M. J. Hannah. The JISCT stereo evaluation. In Proc. of DARPA Image Understanding Workshop, pages 263--274, 1993.

....correspondence problem given in equation (5) assumes that corresponding points have constant brightness. This assumption is quite common in motion or stereo (e.g. 1, 11] but it is often violated in practice. For example, Cox et al. 6] point out that most of the images in the JISCT collection [3] violate the constant brightness assumption. There are several reasons why the constant brightness assumption is invalid. Stereo uses two cameras, and cameras have di#erent internal parameters. The di#erence between two cameras can be modeled as a linear transformation of intensities I = g I # ....

R. Bolles, H. Baker, and M. Hannah. The JISCT stereo evaluation. In DARPA Image Understanding Workshop, pages 263--274, 1993. posteriori estimate for the Ising model can be computed very rapidly using graph cuts [9]. 20

....stereo reconstruction algorithms andshowanexample. 1 Introduction Much discussion has been given over to the need for ways to evaluate image understanding algorithms and systems in an objective and quantitative manner [Sawhney and Hanson, 1990, Weems et al. 1991, Firschein et al. 1993, Bolles et al. 1993] A few calibrated datasets are now available (e.g. see [Willson and Shafer, 1992, Thompson and Owen, 1994, Owen et al. 1996] but reports of either quantitative evaluations of IU methods or the procedures for performing such evaluations on real imagery are still rare. Informal, subjective ....

R. C. Bolles, H. H. Baker, and M. J. Hannah. The JISCT stereo evaluation. In Proc. ARPA Image Understanding Workshop, April 1993.

....is natural to assume that corresponding points in the left and right images have constant brightness. This assumption is quite common in motion or stereo (e.g. 1, 8] but it is often violated in practice. For example, Cox et al. point out in [5] that most of the images in the JISCT collection [3] violate the constant brightness assumption. There are several reasons why the constant brightness assumption is invalid. Stereo uses two cameras, and cameras have di#erent internal parameters. The di#erence between two cameras can be modeled as a linear transformation of intensities I = g I # ....

R. Bolles, H. Baker, and M. Hannah. The JISCT stereo evaluation. In DARPA Image Understanding Workshop, pages 263--274, 1993.

....image. of everyday scenes often contain occlusion regions much larger than those found in popular stereo test imagery. In our lab, common images like Figure 1 contain disparity shifts and occlusion regions over eighty pixels wide. 1 Popular stereo test images, however, like the JISCT test set[ 9], the pentagon image, the white house image, and the Renault part image have maximum occlusion disparity shifts on the order of 20 pixels wide. Regardless of camera configuration, images of the everyday world will have substantially larger occlusion regions than aerial or terrain data. Even ....

....GCPs. Of course, we are limited to how small the occlusion cost can be. If it is smaller than the typical value of correct matches (non zero due to noise) 6 then the algorithm proposes additional occlusion regions such as in path d of Figure 7. For real stereo images (such as the JISCT test set [ 9]) the typical DSI value for incorrectly matched pixels is significantly greater than that of correctly matched ones and performance of the algorithm is not particularly sensitive to the occlusion cost. Also, we note that while we have attempted to remove smoothing influences entirely, there are ....

[Article contains additional citation context not shown here]

R. Bolles, H. Baker, and M. Hannah. The JISCT stereo evaluation. In Proc. Image Understanding Workshop, pages 263-- 274, 1993.

....In many cases, we have not been able to reproduce the published results by using the authors algorithm. This is further complicated by the fact that there is no standard evaluation criterion. PAMI, May 1994 [12] Interesting comparative studies have recently been performed for stereo analysis [4] and optical flow [2] In [2] the authors implemented the nine different techniques they compared. This approach suffers from two drawbacks. First, the amount of work required by the single group is enormous. Second, a re implementation can potentially produce results different from those of the ....

....the nine different techniques they compared. This approach suffers from two drawbacks. First, the amount of work required by the single group is enormous. Second, a re implementation can potentially produce results different from those of the original implementation. By contrast, the authors of [4] designed a test framework and distributed instructions and data to five groups. This approach also has drawbacks. Time and effort are spent coordinating results from the different groups. Additionally, some techniques may have been refined since their last published report. In [4] only three of ....

[Article contains additional citation context not shown here]

R. C. Bolles, H. H. Baker, and M. J. Hannah. The jisct stereo evaluation. In Image Understanding Workshop, pages 263--274, Wash. D.C., 1993.

....with both synthetic and real stereo pairs, and show how our results improve on those of closely related techniques for both accuracy and e ciency. Key words: Computer Vision, Stereo; Depth and Shape Recovery; Area based; Multiple Window. 1 Introduction The aim of computational stereopsis [4, 2] is to reconstruct the 3 D geometry of a scene from two (or more) views, which we call left and right, taken by pinhole cameras. A well known problem is correspondence, i.e. nding which points in the left and right images are projections of the same scene point (a conjugate pair) This is ....

R. C. Bolles, H. H. Baker, and M. J. Hannah. The JISCT stereo evaluation. In Proceedings of the Image Understanding Workshop, pages 263-274, Washington, DC, April 1993. ARPA, Morgan Kaufmann.

.... 2 Institute of Informatics, University of Bern, Switzerland, email jiang or bunke iam.unibe.ch 3 School of Electrical Engineering and Computer Science, Washington State University, Pullman WA 99164 2752, email flynn eecs.wsu.edu been attempted in the areas of optical flow [1] stereo [2], and shape from shading [16] Though these efforts represent a positive step, we feel that a guiding philosophy for the design of a comparative effort is lacking. A collective examination of these works, in addition to our own experience in range image segmentation, suggests that several factors ....

R. C. Bolles, H. H. Baker and M. J. Hannah, "The JISCT Stereo Evaluation", in IUW, Wash. D.C., pp. 263-274, 1993.

....This could help illuminate future research directions. While it seems easy to recognize the need for a comparison of techniques, the problem of going about the comparison is in fact quite difficult. Interesting comparative studies have recently been performed in the areas of stereo analysis [4] and optical flow [2] Each of these offers insights into designing a methodology for a comparative study. However, the methodologies employed in these studies could both be improved. In [2] the authors themselves implemented the nine different techniques they compared. This approach suffers from ....

....First, the amount of work required by the single group is enormous. Second, a re implementation can potentially produce results different from those of the original implementation, perhaps due to errors or differing interpretations of the original description. By contrast, the authors of [4] took the approach of designing a test framework and distributing instructions and data to five research groups, who were then to execute their algorithms and return results for comparison. This approach also has drawbacks. The time spent waiting for all of the participating groups to report ....

[Article contains additional citation context not shown here]

R. C. Bolles, H. H. Baker, and M. J. Hannah. The jisct stereo evaluation. In Image Understanding Workshop, pages 263--274, Wash. D.C., 1993.

No context found.

R. Bolles, H. Baker, and M. Hannah. The JISCT Stereo Evaluation. In DIUW, pages 263--274, 1993.

No context found.

R. C. Bolles, H. H. Baker, and M. J. Hannah. The jisct stereo evaluation. Proceedings of the ARPA Image Understanding Workshop, pages 263--274, 1993.

No context found.

R. C. Bolles, H. H. Baker, and M. J. Hannah. The JISCT stereo evaluation. In Proceedings of the Image Understanding Workshop, pages 263--274, Washington, DC, April 1993. ARPA, Morgan Kaufmann.

No context found.

R. C. Bolles, H. H. Baker, and M. J. Hannah. The JISCT stereo evaluation. In Proceedings of the Image Understanding Workshop, pages 263--274, Washington, DC, April 1993. ARPA, Morgan Kaufmann.

No context found.

R. C. Bolles, H. H. Baker and M. J. Hannah, "The JISCT stereo evaluation," Proc. Image Understanding Workshop, Washington, DC, April 1993, ARPA, Morgan Kaufmann, pp. 263--274.

No context found.

R.C. Bolles and H.H. Baker and M.J. Hannah, 1993, "The JISCT stereo evaluation". Proc. IUW, 263--274.

First 50 documents

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