Shared Parts Latent Topic Model for Image Classification

Abstract:

This paper addresses the problem of accurately classifying image categories without any human interaction. A shared parts latent topic model is presented to share mixture components between categories. Different categories share the similar parts which make the model more accurate. As the number of components is unknown and is to be inferred from the train set, the Dirichlet process is introduced into the model to provide a nonparametric prior for the number of mixture components within each category. Gaussian mixture model is adopted to present the object color feature and the Wishart distribution is applied to estimate the parameters of object shape feature. A number of classification experiments are used to verify the success of our model.

• Periodical: Advanced Materials Research (Volumes 271-273)
• Pages: 1257-1262
• DOI: https://doi.org/10.4028/www.scientific.net/AMR.271-273.1257
• Online since: July 2011
• Authors: Bin Yang
• Keywords: Dirichlet Process, Image Classification, Latent Topic Model, Wishart Distribution

References

[1] F. Monary, D. Gatica-Perez: PLSA-based image auto-annotation: constraining the latent space. Proceedings of the 12th annual ACM international conference on multimedia, pp.348-351, (2004).

DOI: 10.1145/1027527.1027608

[2] D. Blei, A. Ng, M. Jordan: Latent dirichlet allocation. Journal of Machine Learing Research, pp.993-1022, (2003).

[3] C. Dance, J. Williamowski, L. Fan, C. Bray, G. Csurka: Visual categorization with bags of keypoints. ECCV workshop on statistical learning in computer vision, (2004).

[4] F. Jurie, B. Triggs: Creating efficient codebooks for visual recognition. IEEE international conference on computer vision, pp.604-610, (2005).

DOI: 10.1109/iccv.2005.66

[5] F. Li, P. Perona: A bayesian hierarchical model for learning natural scene categories. CVPR, pp.524-531(2005).

[6] C. Li, F. Li: Spatially coherent latent topic model for concurrent object segmentation and classificaton. Proceedings of IEEE International Conferenct on Computer Vision, pp.1-8, (2007).

DOI: 10.1109/iccv.2007.4408965

[7] L. Diane, V. Jakob, J. Frederic: Category level object segmentation by combining bag-of-words models with dirichlet processes and random fields. International Journal Computer Vision, pp.238-253, (2010).

DOI: 10.1007/s11263-009-0245-x

[8] F. Li, R. Fergus, P. Perona: Learning generative visual models from few training examples: An incremental Bayesian approach tested on 101 object categories. CVPR Workshop on Generative Model Based Vision, (2004).

DOI: 10.1109/cvpr.2004.383

[9] A. Torralba, K. P. Murphy, W. T. Freeman: Sharing features: Efficient boosting procedures for multiclass object detection. IEEE Conference on Computer Vision and Pattern Recognition, Vol. 2, pp.762-769, (2004).

DOI: 10.1109/cvpr.2004.1315241

[10] J. Shi, J. Malik: Normalized cuts and image segmentation. IEEE Transation Pattern Anal. Mach. Intell., pp.888-905, (2000).

DOI: 10.1109/34.868688

[11] K. Mikolajczyk, C. Schmid: Scale and affine invariant interest point detectors. International Journal of Computer Vision, pp.63-86, (2004).

DOI: 10.1023/b:visi.0000027790.02288.f2

[12] J. Matas, O. Chum, M. Urban, T. Pajdla: Robust wide baseline stereo from maximally stable extremal regoins. British Machine Vision Conference, pp.384-393, (2002).

DOI: 10.5244/c.16.36

[13] D. G. Lowe: Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision, pp.91-110, (2004).

DOI: 10.1023/b:visi.0000029664.99615.94

[14] R. Neal: Markov chain samling methods for Diriclet process mixture model. Technical Report 9815, University of Toronto, Department of Staitistics, (1998).

[15] A. Gelman, J. B. Carlin, H. S. Stern, D. B. Rubin: Bayesian Data Analysis. Chapman & Hall, (2004).

[16] M. T. Orchard, Bounman: Color quantization of images. IEEE Transations on Signal Processing, pp.2677-2690, (1991).