![](bg5.png)
photos needed. The desired scene should be visible
in the majority of the images. If many persons are
located on the same location in most of the photos, the
required scene is not visible, and the method won’t be
able to generate a plausible result.
As shown in Figure 2, moving backgrounds (like
clouds and trees) won’t necessarily result in bad im-
ages. The result is not the same as in any image, but
still plausible and pleasant looking. This is also valid
for changes in lighting, as can be clearly seen in Fig-
ure 3.
The method works best when the distance to the
desired scene is large and the movement of the cam-
era is limited. This will reduce the effects of par-
allax caused by movements of the camera. Because
the registration uses only linear transformations from
one image to another, no parallax effects can be incor-
porated. If the parallax is large, the registration will
be inaccurate, resulting in a blurred image. Incorpo-
rating parallax remains an open problem until today
(Szeliski, 2006).
We compared our method with other methods. As
can be seen in Figure 6, our method generates more
plausible results compared with the tourist remover
application of Snapmania (Snapmania, 2011). Their
method cannot handle relighting very well.
The group shot application of Agarwala et al.
(Agarwala et al., 2004) provides good results com-
pared with our method, but requires a lot of user input,
which makes it more difficult to use. However, their
method works better when using only 2 or 3 images.
5 CONCLUSIONS
We presented an automatic method to remove tourists,
cars and other moving objects from a set of pho-
tos, while keeping the static scene. Typically, 5 to
10 photos are needed, with a few seconds between
each photo. Alternatively, a video sequence can also
be used. The algorithm uses features to calculate a
transformation between the different images, so that
no tripods or other advanced stabilization devices are
needed during the acquisition of the photos. As can be
seen from the results, the final result is plausible and
pleasant looking. However, the image is not neces-
sarily correct when there are specific background ele-
ments, like clouds or other slow moving objects. This
is not a problem, because the focus of the application
is on the static scene and the generated clouds and
trees are still plausible. The method outperforms ex-
isting techniques when using sufficient input images,
considering the quality of the result and the manual
input required.
ACKNOWLEDGEMENTS
Patrik Goorts would like to thank the IWT for its PhD
specialization bursary.
REFERENCES
Agarwala, A., Dontcheva, M., Agrawala, M., Drucker, S.,
Colburn, A., Curless, B., Salesin, D., and Cohen,
M. (2004). Interactive digital photomontage. In
SIGGraph-04, pages 294–302.
Battiato, S., Gallo, G., Puglisi, G., and Scellato, S. (2007).
Sift features tracking for video stabilization. In Im-
age Analysis and Processing, 2007. ICIAP 2007. 14th
International Conference on, pages 825–830. IEEE.
Bertalmio, M., Sapiro, G., Caselles, V., and Ballester, C.
(2000). Image inpainting. In Proceedings of the 27th
annual conference on Computer graphics and interac-
tive techniques, pages 417–424. ACM Press/Addison-
Wesley Publishing Co.
Censi, A., Fusiello, A., and Roberto, V. (1999). Image sta-
bilization by features tracking. In Image Analysis and
Processing, 1999. Proceedings. International Confer-
ence on, pages 665–667. IEEE.
Chang, J., Hu, W., Cheng, M., and Chang, B. (2002). Digi-
tal image translational and rotational motion stabiliza-
tion using optical flow technique. Consumer Electron-
ics, IEEE Transactions on, 48(1):108–115.
Hartley, R. and Zisserman, A. (2000). Multiple view geom-
etry, volume 6. Cambridge university press.
Hoffman, M. (2010). Automatic tourist remover.
http://www.tipsquirrel.com/index.php/2010/07/autom
atic-tourist-remover/.
Hu, R., Shi, R., Shen, I., and Chen, W. (2007). Video stabi-
lization using scale-invariant features. In Information
Visualization, 2007. IV’07. 11th International Confer-
ence, pages 871–877. IEEE.
Kumar, S., Azartash, H., Biswas, M., and Nguyen, T.
(2011). Real-time affine global motion estimation us-
ing phase correlation and its application for digital im-
age stabilization. Image Processing, IEEE Transac-
tions on, 99:1–1.
Litvin, A., Konrad, J., and Karl, W. (2003). Probabilis-
tic video stabilization using kalman filtering and mo-
saicking. In Proceedings of SPIE Conference on Elec-
tronic Imaging, pages 663–674. Citeseer.
Lowe, D. G. (1999). Object recognition from local scale-
invariant features. In Computer Vision, 1999., vol-
ume 2, pages 1150–1157. Ieee.
Pinto, B. and Anurenjan, P. (2011). Surf based robust video
stabilization using kalman filter. In ICTT Electronics
and Communication.
Rousseeuw, P., Leroy, A., and Wiley, J. (1987). Robust re-
gression and outlier detection, volume 3. Wiley On-
line Library.
Snapmania (2011). Snapmania tourist remover.
http://www.snapmania.com/info/en/trm/.
Szeliski, R. (2006). Image alignment and stitching: A tu-
torial. Foundations and Trends in Computer Graphics
and Vision, 2(1):1–104.
VISAPP 2012 - International Conference on Computer Vision Theory and Applications
196