Discriminative Embedding via Image-to-Class Distances

Xiantong Zhen, Ling Shao and Feng Zheng

In Proceedings British Machine Vision Conference 2014


Image-to-Class (I2C) distance firstly proposed in the naive Bayes nearest neighbour (NBNN) classifier has shown its effectiveness in image classification. However, due to the large nearest-neighbour search, I2C-based methods is extremely time-consuming for testing, especially with high-dimensional local features. In this paper, with the aim to improve and speed up I2C-based methods, we propose a novel discriminative embedding method based on I2C for local feature dimensionality reduction. Our method 1) greatly reduces the computational burden and improves the performance of I2C-based methods after reduction; 2) can well preserve the discriminative ability of local features, thanks to the use of I2C distances; and 3) provides an efficient solution by formulating the objective function as an eigenvector decomposition problem. We apply the proposed method to action recognition showing that it can significantly improve I2C-based classifiers.


Image Classification


Extended Abstract (PDF, 1 page, 250K)
Paper (PDF, 12 pages, 306K)
Bibtex File



Xiantong Zhen, Ling Shao, and Feng Zheng. Discriminative Embedding via Image-to-Class Distances. Proceedings of the British Machine Vision Conference. BMVA Press, September 2014.


	title = {Discriminative Embedding via Image-to-Class Distances},
	author = {Zhen, Xiantong and Shao, Ling and Zheng, Feng},
	year = {2014},
	booktitle = {Proceedings of the British Machine Vision Conference},
	publisher = {BMVA Press},
	editors = {Valstar, Michel and French, Andrew and Pridmore, Tony}
	doi = { http://dx.doi.org/10.5244/C.28.33 }