Supervised spectral subspace clustering for visual dictionary creation in the context of image classification

When building traditional Bag of Visual Words (BOW) for image classification, the K-means algorithm is usually used on a large set of high dimensional local descriptors to build the visual dictionary. However, it is very likely that, to find a good visual vocabulary, only a sub-part of the descriptor space of each visual word is truly relevant. We propose a novel framework for creating the visual dictionary based on a spectral subspace clustering method instead of the traditional K-means algorithm. A strategy for adding supervised information during the subspace clustering process is formulated to obtain more discriminative visual words. Experimental results on real world image dataset show that the proposed framework for dictionary creation improves the classification accuracy compared to using traditionally built BOW.