Details
| Original language | English |
|---|---|
| Publication status | Published - 2011 |
| Externally published | Yes |
| Event | TREC Video Retrieval Evaluation, TRECVID 2011 - Gaithersburg, MD, United States Duration: 5 Dec 2011 → 7 Dec 2011 |
Conference
| Conference | TREC Video Retrieval Evaluation, TRECVID 2011 |
|---|---|
| Country/Territory | United States |
| City | Gaithersburg, MD |
| Period | 5 Dec 2011 → 7 Dec 2011 |
Abstract
In this paper, we summarize our results for the semantic indexing task at TRECVID 2011. Last year, we showed that the use of object detection results as additional midlevel features improved the overall performance of a bag-of-visual-words (BoVW) approach. This year, we repeated the experiment on a large concept vocabulary of 346 classes. In addition, we investigated whether feature descriptions of object regions can also improve the concept detection performance. Due to the large number of face-related concepts, like "adult", "female", "male", "dark skinned person", "first lady", "glasses", or "arafat", BoVW features are extracted from face regions and are used as an additional feature representation. Furthermore, a new post-processing scheme is introduced, that leads to a rescoring of shots based on concept relations. The experiments showed that the use of additional object-based features significantly improved the concept detection performance. Further improvements are attained using region-based BoVW features and relation-based rescoring. Altogether, our best run achieved a mean inferred average precision of 12.3% and we submitted the best results for the concepts "overlaid text" and "two persons".
ASJC Scopus subject areas
- Computer Science(all)
- Computer Graphics and Computer-Aided Design
- Computer Science(all)
- Computer Vision and Pattern Recognition
- Computer Science(all)
- Human-Computer Interaction
- Computer Science(all)
- Software
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
2011. Paper presented at TREC Video Retrieval Evaluation, TRECVID 2011, Gaithersburg, MD, United States.
Research output: Contribution to conference › Paper › Research › peer review
}
TY - CONF
T1 - University of Marburg at TRECVID 2011
T2 - TREC Video Retrieval Evaluation, TRECVID 2011
AU - Mühling, Markus
AU - Ballafkir, Khalid
AU - Ewerth, Ralph
AU - Freisleben, Bernd
PY - 2011
Y1 - 2011
N2 - In this paper, we summarize our results for the semantic indexing task at TRECVID 2011. Last year, we showed that the use of object detection results as additional midlevel features improved the overall performance of a bag-of-visual-words (BoVW) approach. This year, we repeated the experiment on a large concept vocabulary of 346 classes. In addition, we investigated whether feature descriptions of object regions can also improve the concept detection performance. Due to the large number of face-related concepts, like "adult", "female", "male", "dark skinned person", "first lady", "glasses", or "arafat", BoVW features are extracted from face regions and are used as an additional feature representation. Furthermore, a new post-processing scheme is introduced, that leads to a rescoring of shots based on concept relations. The experiments showed that the use of additional object-based features significantly improved the concept detection performance. Further improvements are attained using region-based BoVW features and relation-based rescoring. Altogether, our best run achieved a mean inferred average precision of 12.3% and we submitted the best results for the concepts "overlaid text" and "two persons".
AB - In this paper, we summarize our results for the semantic indexing task at TRECVID 2011. Last year, we showed that the use of object detection results as additional midlevel features improved the overall performance of a bag-of-visual-words (BoVW) approach. This year, we repeated the experiment on a large concept vocabulary of 346 classes. In addition, we investigated whether feature descriptions of object regions can also improve the concept detection performance. Due to the large number of face-related concepts, like "adult", "female", "male", "dark skinned person", "first lady", "glasses", or "arafat", BoVW features are extracted from face regions and are used as an additional feature representation. Furthermore, a new post-processing scheme is introduced, that leads to a rescoring of shots based on concept relations. The experiments showed that the use of additional object-based features significantly improved the concept detection performance. Further improvements are attained using region-based BoVW features and relation-based rescoring. Altogether, our best run achieved a mean inferred average precision of 12.3% and we submitted the best results for the concepts "overlaid text" and "two persons".
UR - http://www.scopus.com/inward/record.url?scp=84905234447&partnerID=8YFLogxK
M3 - Paper
Y2 - 5 December 2011 through 7 December 2011
ER -