Image Retrieval with Simple Invariant Features Based Hierarchical Uniform Segmentation

Abstract

According to local information of images, region- based image retrieval is the focus of recent research works, as the approaches based global features can not achieve the expectation querying results. The objects of interest generally occupy only one small part of images, so the image segmentation with different object regions must be conducted for the region-based image retrieval schemes. However, accurate object segmentation is still beyond current computer vision technique. Here, we proposed one feasible image retrieval scheme based the hierarchical uniform segmentations, which avoid the complexity of image segmentations. Firstly, the querying image is segmented into equal blocks at different hierarchical levels, and the more blocks with larger hierarchical levels. Then, according to the similar metrics of these different size blocks to the expectation image into segmentations, the images containing querying objects can be retrieved with information about scales and locations of query objects in retrieved images. Finally, the proposed image retrieval schemes are tested by experiments via database with 500 images, and the retrieval accuracy can achieve 78% for the optimal similar metric threshold, and is comparable to that of region-based schemes.