A Novel Approach to Remote Sensing Image Retrieval with Multi-feature VP-Tree Indexing and Online Feature Selection

Abstract

With the development of remote sensing (RS) techniques, the amount of RS images increases dramatically. It is a challenge to utilize those RS big data efficiently. Content-based Image Retrieval (CBIR) is a typical approximate similarity search problem, which needs to establish an effective index structure to reduce the time of retrieval. By analyzing the limitations of commonly-used indexing mechanisms in the current CBIR system, we propose a novel scheme that dynamically combines vantage point tree (vp-tree) indexes to CBIR by using spacing-correlation strategies to determine the vantage points. Borrowing ideas from feature selection, we have also put forward a new measure to adaptively online select proper vp-tree indexing in different feature spaces, the distance-contrast-based indexing validity index (DCIVI). And we then employ vp-tree index structure in each feature space, which can properly describe the content of the RS image by the chosen features. Experimental results on various typical land covers retrieval validate that the proposed method is effective and not only is the response speed increased by 70 100 times, but also the retrieval quality (in terms of precision and recall) is improved.

Publication
2016 IEEE Second International Conference on Multimedia Big Data (BigMM)
Lixin Yuan
Lixin Yuan
Lecturer