摘要：To narrow down the gap between low-level visual features and high-level semantic concepts in content-based image retrieval(CBIR) system, a new dimensionality reduction algorithm called tensor biased discriminant Euclidean embedding (TBDEE) is proposed in this paper. The key idea of this algorithm is as follows: First, the image data are represented with high order tensor structure so that the correlations among pixels within different dimensions can be effectively utilized. Second, the optimal objective function is constructed by considering both the intraclass geometry and interclass discrimination. Third, the transformation matrices are obtained by iteratively unfolding the tensor along different tensor directions. Finally, the high-dimensional image data are projected into the lower-dimensional feature for image retrieval by using the obtained transformation matrices. Comprehensive experiments on the COREL database demonstrate that the proposed algorithm outperforms other related algorithms in terms of effectiveness and efficiency.
