Efficient image search system with multi-scale database using profile-based pre-selection and coarse matching

Abstract

Artifact-metrics technology is gaining more research interests, along with the expansion of its applications. One of its challenges is an efficient image search, in which a match is to be identified in a large multi-scale image database with a given, possibly distorted, query image having unknown location, orientation, and scale. To approach this computational efficiency problem of image database search, we conducted a preliminary feasibility study focused specifically on aerial photo search problem. We propose a highly efficient image search system to find a match in the multi-layered database of images with multiple magnitudes. The system first pre-selects matching candidates based on comparison results of image profiles such as frequency spectra, so that the following matching stages focus on the appropriate scale layer to accelerate search. Then in the coarse matching stage, the down-sampled query image is compared with images in a lower-magnitude layer using a scale-invariant matcher based on local feature descriptors. This paper outlines our interim proposed approach and discusses its feasibility and performance based on the experimental results from our ongoing research work.