![](bg6.png)
classification of segmented but unknown objects. In
this approach, state-of-the-art texture and color
features are extracted from regions that cover the
entire object with and without background-
modifications. Results of an extensive evaluation
indicate that the proposed approach offers the
opportunity to improve the task of object class
detection in combination with efficient segmentation
approaches. The experiments of this work
investigated perfect segmentations as well as
inaccurate ones. The classification was done with a
nearest neighbor matching strategy and different
dissimilarity measures to keep the evaluation as
simple and universally valid as possible.
In the evaluation, we have first shown that it
does matter how the regions of segmented objects
are prepared for semi-local feature extraction.
Regions where the object and its background are
modified can improve the overall classification rate
significantly compared to unmodified regions,
especially for accurate segmentations. Secondly,
square bounding boxes achieves better results than
tight, rectangular bounding boxes. Thirdly, texture
features perform better than color features and
improvements of a few percent can be achieved
when the right dissimilarity measures are chosen.
The Jeffrey divergence and Chi-Square correlation
performed best for all feature types and region
preparation techniques. We conclude that semi-local
features are good candidates to improve object
detection systems due to their simplicity and the
promising results in this work. Furthermore, we plan
to investigate semi-local features in an integrated
object detection system to verify this assumption.
ACKNOWLEDGEMENTS
The author would like to thank Horst Eidenberger
for his feedback and support. The research leading to
this publication has received funding from the
Austrian FIT-IT project ‘IV-ART – Intelligent
Video Annotation and Retrieval Techniques’.
REFERENCES
Carreira, J., Sminchisescu, C., 2010. Constrained
parametric min cuts for automatic object
segmentation. CVPR.
Comaniciu, D., Meer, P., 2002. Mean shift: A robust
approach toward feature space analysis. PAMI.
Csurka, G., Perronnin, F., 2010. An efficient approach to
semantic segmentation. IJCV.
Dalal, N., Triggs, B., 2005. Histograms of oriented
gradients for human detection. CVPR.
Everingham, M., Van Gool, L., Williams, C., Winn, J.,
Zisserman, A., 2010. The PASCAL Visual Object
Classes (VOC) challenge. IJCV.
Frigo, M., Johnson, S., 2005. The design and implement-
tation of FFTW3. Proc. Program Generation, Optimi-
zation, and Platform Adaptation
Hoiem, D., Efros, A., Hebert, M., 2005. Geometric
context from a single image. ICCV.
Hoiem, D., Stein, A., Efros, A., Hebert, M., 2011.
Recovering occlusion boundaries. IJCV.
Lampert, C., Blaschko, M., Hofmann, T., 2008. Beyond
sliding windows: Object localization by efficient
subwindow search. CVPR.
Lazebnik, S., Schmid, C., Ponce, J., 2006. Beyond bags of
features: Spatial pyramid matching for recognizing
natural scene categories. CVPR.
Leibe, B., Leonardis, A., Schiele, B., 2008. Robust object
detection with interleaved categorization and
segmentation. IJCV.
Li, F., Carreira, J., Sminchisescu, C., 2007. Object
recognition as ranking holistic figure-ground
hypotheses. CVPR.
Liu, H., Song, D., Rüger, S., Hu, S., Uren, V., 2008.
Comparing dissimilarity measures for content-based
image retrieval. AIRS.
Lowe, D., 2004. Distinctive image features from scale-
invariant keypoints. IJCV.
Manjunath, B., Ohm, J.-R., Vasudevan, V., Yamada, A.,
2001. Color and texture descriptors. Trans. on Circuits
and Systems for Video Technology.
Mikolajczyk K., Schmid, C., 2005a. A performance
evaluation of local descriptors. Trans. PAMI.
Mikolajczyk, K., Tuytelaars, T., Schmid, C., Zisserman,
A., Matas, J., Schaffalitzky, F., Kadir, T., Van Gool,
L., 2005b. A comparison of affine region detectors.
IJCV.
Oliva, A., Torralba, A., 2006. Building the GIST of a
Scene: The Role of Global Image Features in
Recognition, Visual Perception, Progress in Brain
Research.
Pantofaru, C., Schmid C., Hebert, M., 2008. Object reco-
gnition by integrating multiple image segmentations.
ECCV
.
Rabinovich A., Vedaldi, A., Belongie, S., 2007. Does ima-
ge segmentation improve object categorization? Tech.
Rep. CS2007-090.
Russell, B., Freeman, W., Efros, A., Sivic, J., Zisserman,
A., 2006. Using multiple segmentations to discover
objects and their extent in image collections. CVPR.
Shi, J., Malik, J., 1997. Normalized cuts and image
segmentation. CVPR.
Toshev, A., Taskar, B., Daniilidis, K, 2010. Object
detection via boundary structure segmentation. CVPR.
Van de Sande, K., Gevers, T., Snoek, C., 2010. Evaluating
color descriptors for object and scene recognition.
PAMI.
SEMI-LOCAL FEATURES FOR THE CLASSIFICATION OF SEGMENTED OBJECTS
175