ABSTRACT
Visual concept learning often requires a large set of training images. In practice, nevertheless, acquiring noise-free training labels with sufficient positive examples is always expensive. A plausible solution for training data collection is by sampling the largely available user-tagged images from social media websites. With the general belief that the probability of correct tagging is higher than that of incorrect tagging, such a solution often sounds feasible, though is not without challenges. First, user-tags can be subjective and, to certain extent, are ambiguous. For instance, an image tagged with "whales" may be simply a picture about ocean museum. Learning concept "whales" with such training samples will not be effective. Second, user-tags can be overly abbreviated. For instance, an image about concept "wedding" may be tagged with "love" or simply the couple's names. As a result, crawling sufficient positive training examples is difficult. This paper empirically studies the impact of exploiting the tagged images towards concept learning, investigating the issue of how the quality of pseudo training images affects concept detection performance. In addition, we propose a simple approach, named semantic field, for predicting the relevance between a target concept and the tag list associated with the images. Specifically, the relevance is determined through concept-tag co-occurrence by exploring external sources such as WordNet and Wikipedia. The proposed approach is shown to be effective in selecting pseudo training examples, exhibiting better performance in concept learning than other approaches such as those based on keyword sampling and tag voting.
- M. Ames and M. Naaman. Why we tag: Motivations for annotation in mobile and online media. In ACM SIGCHI, 2007. Google ScholarDigital Library
- A. Ulges et al. Learning automatic concept detectors from online video. Comput. Vis. Image Understand, 2009. Google ScholarDigital Library
- D. Liu et al. Tag ranking. In ACM WWW, 2009. Google ScholarDigital Library
- G.-J. Qi et al. Transductive inference with hierarchical clustering for video annotation. In ICME, 2007.Google ScholarCross Ref
- J. Tang et al. Inferring semantic concepts from community-contributed images and noisy tags. In ACM MM, 2009. Google ScholarDigital Library
- K. Bischoff et al. Can all tags be used for search. In ACM CIKM, 2008. Google ScholarDigital Library
- Q. Tian et al. A new analysis of the value of unlabeled data in semi-supervised learning for image retrieval. In ICME, 2004.Google Scholar
- S.-F. Chang et al. Columbia University/VIREO-CityU/IRIT TRECVID 2008 high-level feature extraction and interactive video search. In TRECVID, 2008.Google Scholar
- T.-S. Chua et al. NUS-WIDE: A real-world web image database from national university of singapore. In CIVR, 2009. Google ScholarDigital Library
- Y.-G. Jiang et al. Representations of keypoint-based semantic concept detection: A comprehensive study. IEEE Trans. on Multimedia, 12(1):42--53, 2010. Google ScholarDigital Library
- Y.-G. Jiang, C.-W Ngo, and J. Yang. Towards optimal bag-of-features for object categorization and semantic video retrieval. In CIVR, 2007. Google ScholarDigital Library
- D. Jurafsky and J. H. Martin. Speech and language processing. Prentice-Hall, 2000. Google ScholarDigital Library
- L. S. Kennedy, S.-F. Chang, and I. V. Kozintsev. To search or to label. In ACM MIR, 2006. Google ScholarDigital Library
- L.-J. Li and L. Fei-Fei. OPTIMOL: automatic object picture collection via incremental model learning. Int. J. of Computer Vision, 2009. Google ScholarDigital Library
- X.-R. Li and C. G. M. Snoek. Visual categorization with negative examples for free. In ACM MM, 2009. Google ScholarDigital Library
- X.-R. Li, C. G. M. Snoek, and M. Worring. Learning social tag relevance by neighbor voting. IEEE Trans. on MM, 11(7):1310--1322, 2009. Google ScholarDigital Library
- M. R. Naphade and J. R. Smith. On the detection of semantic concepts at TRECVID. In ACM MM, 2004. Google ScholarDigital Library
- G. Quénot and S. Ayache. TRECVID 2009 collaborative annotation. http://mrim.imag.fr/tvca/.Google Scholar
- F. Schroff, A. Criminisi, and A. Zisserman. Harvesting image databases from the web. In ICCV, 2007.Google ScholarCross Ref
- A. T. Setz and C. G. M. Snoek. Can social tagged images aid concept-based video search. In ICME, 2009. Google ScholarDigital Library
- S. Tong and E. Chang. Support vector machine active learning for image retrieval. In ACM MM, 2001. Google ScholarDigital Library
- G. Wang, T.-S. Chua, and M. Zhao. Exploring knowledge of sub-domain in a multi-resoluation bootstrapping framework for concept detection in news. In ACM MM, 2008. Google ScholarDigital Library
- R. Yan, A. G. Hauptmann, and R. Jin. Negative pseudo-relevance feedback in content-based video retrieval. In ACM MM, 2003. Google ScholarDigital Library
- R. Yan and M. R. Naphade. Semi-supervised cross feature learning for semantic concept detection in video. In CVPR, 2005. Google ScholarDigital Library
Index Terms
- On the sampling of web images for learning visual concept classifiers
Recommendations
Sampling and Ontologically Pooling Web Images for Visual Concept Learning
Part 1Sufficient training examples are essential for effective learning of semantic visual concepts. In practice, however, acquiring noise-free training examples has always been expensive. Recently the rapid popularization of social media websites, such as ...
Learning automatic concept detectors from online video
Concept detection is targeted at automatically labeling video content with semantic concepts appearing in it, like objects, locations, or activities. While concept detectors have become key components in many research prototypes for content-based video ...
Improving multi-view semi-supervised learning with agreement-based sampling
Combined Learning Methods and Mining Complex DataSemi-supervised learning algorithms are widely used to build strong learning models when there are not enough labeled instances. Some semi-supervised learning algorithms, including co-training and co-EM, use multiple views to build learning models. Past ...
Comments