Index model for image retrieval using SIFT distortion

Abstract

An image index model had been proposed that uses the distortion produced by using vector quantisation (VQ) on block vectors of the image to index the image database. In this paper, the novel SIFT distortion (SD) is used as a similarity measure to enable faster retrieval of images from the index model. Each image is represented as a single distortion value. The SD value is compared with various other similarity measures. The results show higher precision for the SD values for the same recall. SD is scale invariant and hence has higher mean average precision for the retrieval of similar images.