Effectiveness of the nearest neighbor's significance in the similarity retrieval of images

摘要

Nearest-neighbor (NN) search in high dimensional space is widely used for the similarity retrieval of images. Recent research results in the literature reveal that NN-search might return insignificant NNs in high dimensional space because points could be so scattered that every distance between them might yield no significant difference. Hence. we devised a way to estimate the significance of NNs based on the local intrinsic dimensionality. In this paper, we evaluate the proposed method with applying it to the similarity retrieval of natural photo images. The experimental result demonstrates the effectiveness of the proposed method.