Robust image retrieval using multiview scalable vocabulary trees

Content-based image retrieval using a Scalable Vocabulary Tree (SVT) built from local scale-invariant featuresis an eective method of fast search through a database. An SVT built from fronto-parallel database images,however, is ineective at classifying query images that suer from perspective distortion. In this paper, wepropose an ecient server-side extension of the single-view SVT to a set of multiview SVTs that may be simulta-neously employed for image classi“cation. Our solution results in signi“cantly better retrieval performance whenperspective distortion is present. We also develop an analysis of how perspective increases the distance betweenmatching query-database feature descriptors.Keywords: image retrieval, feature matching, scalable vocabulary tree, perspective distortion 1. INTRODUCTION One successful approach to content-based image retrieval in recent years relies on robust local features. The Scale-Invariant Feature Transform (SIFT) extracts features from an image that can survive moderate geometric andillumination distortions.