Learning Sparse Covariance Patterns for Natural Scenes
Rehg, James M.
MetadataShow full item record
For scene classification, patch-level linear features do not always work as well as hand-crafted features. In this paper, we present a new model to greatly improve the discrimination power of linear features in classification by introducing covariance patterns. We analyze the properties of covariance, along with their fundamental importance, and present a generative model to properly utilize them. With this set of covariance information, in our framework, even the most naive linear features that originally lack the vital ability in classification become powerful. Experiments show that the performance of our new covariance model based on linear features is comparable with or even better than hand-crafted features in scene classification.