A Novel Hierarchical Image Retrieval Based Geometric Invariants and Normalized Histogram
Mingxin Zhang, Lu Zhaogan, Junyi Shen · Third International Conference on Semantics, Knowledge and Grid (SKG 2007) · 2007
Image retrieval is generally implemented by image matching or based-regions retrieval, but it's difficult to balance retrieval performance and complexity. Query images may appear with different scales and rotations in different images, so a hierarchical image segmentation is proposed to partition the retrieved images into equal blocks with different sizes at different levels. Then, the similar metrics of these sub- blocks to query image, are evaluated to retrieve those sub-blocks with contents in query images. Meanwhile, information about scales and locations of query objects in retrieved images can also be returned. The hierarchical block image retrieval schemes with geometric invariants, normalized histograms and their combinations are tested by experiments via a database with 500 images, respectively. The retrieval accuracy with geometric invariants as invariant features can achieve 78% for the optimal similar metric threshold. Furthermore, the scheme can also work with different size images.