Analysis of the Multi-Dimensional Scale Saliency Algorithm and Its Application to Texture Categorization
A new approach for multi-dimensional Scale Saliency (MDSS) was lately introduced. In this approach, the Scale Saliency algorithm by Kadir and Brady is extended to the multi-dimensional domain. The MDSS algorithm is based on alternative entropy and divergence estimation methods whose complexity does not increase exponentially with data dimensionality. However, MDSS has not been applied to any practical problem yet. In this paper we apply the MDSS algorithm to the texture categorization problem, and we provide further experiments in order to assess the suitability of different estimators to the algorithm. We also propose a new divergence measure based on the k-d partition algorithm.
KeywordsTexture Categorization Query Image Data Dimensionality Partition Algorithm Entropy Estimation
- 3.Mikolajczyk, M., Leibe, B., Schiele, B.: Local features for object class recognition. In: Proc 10th IEEE Int. Conf. Comp. Vision, vol. 2, pp. 1792–1799 (2005)Google Scholar
- 5.Newman, P., Ho, K.: SLAM-loop closing with visually salient features. In: Proc. IEEE Int. Conf. Rob. Aut., pp. 635–642 (2005)Google Scholar
- 6.Suau, P., Escolano, F.: Multi-dimensional scale saliency feature extraction based on entropic graphs. In: Proc. of the 4th Int. Symp. on Visual Computing, vol. 2, pp. 170–180 (2008)Google Scholar
- 7.Suau, P., Escolano, F.: A new feasible approach to multi-dimensional scale saliency. In: Proc. 11th Int. Conf. on Advanced Concepts for Intel. Vision Systems, pp. 77–88 (2009)Google Scholar
- 12.Lowe, D.: Object recognition from local scale-invariant features. In: Proc. of the 7th IEEE Int. Conf. on Comp. Vision, vol. 2, pp. 1150–1157 (1999)Google Scholar