Scene modeling in global-local view for scene classification
Aiwen Jiang, Chunheng Wang, Baihua Xiao · 2008
Scene classification aims to automatically label an image among a set of semantic categories. The issue of scene modeling is critical to its classification performance. Inspired by recent psychology progresses on visual perception, we unify the current popular strategies into a 'gist' framework, and suggest a global-local view to model scenes. We evaluate our strategy on the 13 class scenes dataset mostly cited. The experiment results show that our method significantly outperforms the state-of-art methods. We believe it will give a fresh look at how to effectively model scene to benefit for scene analysis.