Visually significant feature point maps for image retrieval applications

In this paper, we propose a hybrid feature point detector based on a fusion of a multi-scale edge map and a wavelet based interest point representation. The feature points thus obtained are used for efficient image indexing and retrieval. Proposed method preserves an optimum number of significant feature points. Our method is compared against other existing interest point detectors. Efficacy of the proposed method is evaluated via a image retrieval system against the standard test databases. A similarity distance measure like Hausdorff distance is used for image matching, indexing and ranking of results considering the spatial information of the local structures in the image.