Cross-Codebook Image Classification
Representing images by bag of visual codes (BoVC) features has been the cornerstone of state-of-the-art image classification system. Since the BoVC features depend on a precomputed codebook in use, when the codebook applied to test images differs from the codebook of an existing image classification system, the system becomes inapplicable. To resolve the codebook incompatibility problem, we propose in this paper cross-codebook image classification. This is achieved by transforming BoVC features derived from one codebook to make them compatible with another codebook. Two BoVC transform methods, i.e., code-reassignment and least squares, are studied. Experiments on a popular image classification benchmark set show that both methods are better than random guess when crossing the codebooks. In particular, when the BoVC features are transformed from a higher dimension to a relatively small dimension, cross-codebook image classification has a similar performance compared to within-codebook image classification, with a relative performance loss of 1.3% only. The results justify the feasibility of the proposed cross-codebook image classification.
KeywordsImage classification bag of visual codes cross-codebook
Unable to display preview. Download preview PDF.
- 1.Perronnin, F., Akata, Z., Harchaoui, Z., Schmid, C.: Towards good practice in large-scale learning for image classification. In: CVPR, pp. 3482–3489 (2012)Google Scholar
- 4.Csurka, G., Dance, C., Fan, L., Willamowski, J., Bray, C.: Visual categorization with bags of keypoints. In: ECCV Workshop on Statistical Learning in Computer Vision, vol. 1, p. 22 (2004)Google Scholar
- 6.Chua, T.S., Tang, J., Hong, R., Li, H., Luo, Z., Zheng, Y., Zheng, Y.: Nus-wide: a real-world web image database from national university of singapore. In: CIVR, p. 48 (2009)Google Scholar
- 7.Hastie, T., Tibshirani, R., Friedman, J.: The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer (2001)Google Scholar
- 8.Nowak, S., Huiskes, M.: New strategies for image annotation: Overview of the photo annotation task at Image CLEF 2010. In: CLEF (2010)Google Scholar
- 9.Maji, S., Berg, A.C., Malik, J.: Classification using intersection kernel support vector machines is efficient. In: CVPR, pp. 1–8. IEEE (2008)Google Scholar