Multi-feature Image Retrieval by Nonlinear Dimensionality Reduction

Multi-feature fusion is effective in raising the matching performance of image retrieval. However, the "Curse of Dimensionality" has to be solved. Traditional dimensionality reduction methods cannot reflect the high-order correlation among features and are not accommodated for the small sample size problem. In this paper, we propose a new dimensionality reduction algorithm by fusing Kernel Principal Component Analysis (KPCA) and Nonparametric Discriminant Analysis (NDA). Firstly KPCA is used to compress the dimensionality for the small sample set and then NDA is added to raise the separability of features in the derived subspace. The proposed method is tested on the Corel image set, in which the color, texture and shape features are combined and compressed to test the performance of retrieval. The experimental results show a better precision-recall curve (PVR) than those state-of-the-art methods.