A pyramid nearest neighbor search kernel for object categorization
Hong Cheng, Rongchao Yu, Zicheng Liu, Yiguang Liu · 2012
Nearest-Neighbor based Image Classification (N-NIC) has drawn considerable attention in the past sev-eral years because it does not require classifier training. Similar to an orderless Bag-of-Feature image represen-tation, the traditional NNIC ignores global geometric correspondence. In this paper, we present a technique to exploit the global geometric correspondence in a n-earest neighbor classifier framework. We divide an im-age into increasingly fine sub-regions like the Spatial Pyramid Matching (SPM) approach, and introduce a Pyramid Nearest Neighbor Search kernel by measuring the search similarity between a local descriptor and a feature set in each pyramid window. Instead of using a fixed weighting as in SPM, the weights of the pyra-mid windows are learned in a class-dependent manner. By doing so, we learn a class-specific geometric corre-spondence. Finally, an optimal nearest neighbor clas-sifier framework is developed to incorporate the kernel functions over different pyramid windows. We evaluate our proposed approach on a number of public dataset-s, and show the results significantly outperform existing techniques. 1.