A supervised LPP and neural network based scene classification with color histogram and camera metadata (コンピュータビジョンとイメージメディア)

概要

論文の詳細を見る
Scene classification (e.g., landscape, sunset, night-landscape, etc.) is still a challenging problem in computer vision. Scene classification based only on low-level vision cues has had limited success on unconstrained image sets. In other hand, camera metadata related to capture conditions provides cues independent of the captured scene content that can be used to improve classification performance. Analysis of camera metadata statistics for images of each class revealed that some metadata fields are most discriminative for some classes. So, in this paper, we proposed to use the combined feature of scene color histogram and camera metadata, and then using supervised Locality preserving projection (LPP) for feature space transformation and dimension reduction, and finally, adapt Probabilistic neural network for scene classification. Experimental results show that the classification accuracy rate can be improved compared with using PCA (Principal Component Analysis) subspace learning method, and are also better than that with only the low-level vision feature (color histogram).
2009-03-06