Local Dependencies for near-duplicate retrieva

Abstract

Image representation plays an important part in many real-world multimedia ap- plications. Over the past few years, great progresses have been achieved in this area. For example, Bag-of-Words (BOW) model is widely used in object recognition and image retrieval. However, these models do not consider the spatial information suffi- ciently and have severely limited descriptive ability. It is obvious the both objects and scenes have their own structures. Treating local features as independent factors will lose information, thus result in bad results. Initial researches have shown that spatial constrains play an important role in a good model for image recognition and retrieval.