Binary Stereo Matching
Binary Stereo Matching
Binary Stereo Matching
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Average Error (%)<br />
7<br />
6.5<br />
6<br />
5.5<br />
5<br />
512 1024 2048 4096<br />
Descriptor Length(bits)<br />
(a) Quality<br />
Average Running Time(s)<br />
50<br />
40<br />
30<br />
20<br />
10<br />
0<br />
512 1024 2048 4096<br />
Descriptor Length(bits)<br />
(b) Speed<br />
Figure 3. The influence of descriptor<br />
length on the performance of BSM.<br />
Also, the descriptor length affects depth map’s quality<br />
because longer descriptor implies a dense sampling. To<br />
show the influence of descriptor length on the performance<br />
of BSM, we test BSM with different n on the<br />
traditional four datasets from Middlebury. Experimental<br />
result is shown in Figure 3, which is consistent with<br />
the analysis above. This interesting property of BSM<br />
makes it easy to gain different tradeoff between speed<br />
and quality in different scenarios.<br />
5. Conclusion<br />
In this paper, a novel cost computation and aggregation<br />
approach for stereo matching is proposed.<br />
Combining our cost computation and aggregation approach<br />
with WTA strategy, we design a new local stereo<br />
method called binary stereo matching. The proposed algorithm<br />
is mainly based on binary and integer computations,<br />
so it is fast and fits for embedded or mobile devices.<br />
Experimental results show that BSM has a better<br />
performance either on traditional stereo datasets or on<br />
new datasets with radiometric differences. In the future,<br />
we will definitely incorporate our cost computation and<br />
aggregation approach into global optimization and implement<br />
BSM on GPU to achieve real-time matching.<br />
6. Acknowledgement<br />
This work has been partially supported by the<br />
Development Plan of China (973) under Grant No.<br />
2011CB302206, the National Natural Science Foundation<br />
of China under Grant No. 60833009/60933013,<br />
and the Research Grant of Tsinghua-Tencent Joint Lab.<br />
References<br />
[1] C. Barnes, E. Shechtman, A. Finkelstein, and D. B. Goldman.<br />
Patchmatch: a randomized correspondence algorithm<br />
for structural image editing. In ACM SIGGRAPH<br />
2009 papers, SIGGRAPH ’09, pages 24:1–24:11, New<br />
York, NY, USA, 2009. ACM.<br />
[2] M. Calonder, V. Lepetit, C. Strecha, and P. Fua. BRIEF:<br />
<strong>Binary</strong> Robust Independent Elementary Features. In European<br />
Conference on Computer Vision, pages 778–792,<br />
2010.<br />
[3] L. De-Maeztu, S. Mattoccia, A. Villanueva, and<br />
R. Cabeza. Linear stereo matching. In Computer Vision<br />
(ICCV), 2011 IEEE International Conference on, pages<br />
1708 –1715, nov. 2011.<br />
[4] M. Gong, R. Yang, L. Wang, and M. Gong. A performance<br />
study on different cost aggregation approaches<br />
used in real-time stereo matching. International Journal<br />
of Computer Vision (IJCV, 2007.<br />
[5] K. He, J. Sun, and X. Tang. Guided image filtering. In<br />
Proceedings of the 11th European conference on Computer<br />
vision: Part I, ECCV’10, pages 1–14, Berlin, Heidelberg,<br />
2010. Springer-Verlag.<br />
[6] H. Hirschmuller and D. Scharstein. Evaluation of cost<br />
functions for stereo matching. In Computer Vision and<br />
Pattern Recognition, 2007. CVPR ’07. IEEE Conference<br />
on, pages 1 –8, june 2007.<br />
[7] A. Hosni, M. Bleyer, M. Gelautz, and C. Rhemann. Local<br />
stereo matching using geodesic support weights. In Image<br />
Processing (ICIP), 2009 16th IEEE International Conference<br />
on, pages 2093 –2096, nov. 2009.<br />
[8] C. R. Michael Bleyer and C. Rother. Patchmatch stereo<br />
- stereo matching with slanted support windows. In Proceedings<br />
of the British Machine Vision Conference, pages<br />
14.1–14.11. BMVA Press, 2011.<br />
[9] C. Rhemann, A. Hosni, M. Bleyer, C. Rother, and<br />
M. Gelautz. Fast cost-volume filtering for visual correspondence<br />
and beyond. In Computer Vision and Pattern<br />
Recognition (CVPR), 2011 IEEE Conference on, pages<br />
3017 –3024, june 2011.<br />
[10] D. Scharstein and R. Szeliski. Middlebury stereo vision<br />
website. http://vision.middlebury.edu/stereo/.<br />
[11] D. Scharstein and R. Szeliski. A taxonomy and evaluation<br />
of dense two-frame stereo correspondence algorithms.<br />
Int. J. Comput. Vision, 47:7–42, April 2002.<br />
[12] C. Tomasi and R. Manduchi. Bilateral filtering for gray<br />
and color images. In Proceedings of the Sixth International<br />
Conference on Computer Vision, ICCV ’98, pages<br />
839–, Washington, DC, USA, 1998. IEEE Computer Society.<br />
[13] F. Tombari, S. Mattoccia, L. Di Stefano, and E. Addimanda.<br />
Classification and evaluation of cost aggregation<br />
methods for stereo correspondence. In Computer Vision<br />
and Pattern Recognition, 2008. CVPR 2008. IEEE Conference<br />
on, pages 1 –8, june 2008.<br />
[14] K.-J. Yoon and I. S. Kweon. Adaptive support-weight<br />
approach for correspondence search. Pattern Analysis and<br />
Machine Intelligence, IEEE Transactions on, 28(4):650 –<br />
656, april 2006.<br />
[15] R. Zabih and J. Woodfill. Non-parametric local transforms<br />
for computing visual correspondence. In Proceedings<br />
of the third European conference on Computer Vision<br />
(Vol. II), pages 151–158, Secaucus, NJ, USA, 1994.<br />
Springer-Verlag New York, Inc.<br />
359