This paper addresses the problem of object-based image retrieval, by using local feature extraction and a relevance feedback mechanism for quickly narrowing down the image search process to the user needs. This approach relies on the hypothesis that semantically similar images are clustered in some feature space and, in this scenario: (i) computes image signatures that are invariant to scale and rotation using SIFT, (ii) calculates the vector of locally aggregated descriptors (VLAD) to make a fixed length descriptor for the images, (iii) reduce the VLAD descriptor dimensionality with Principal Component Analysis (PCA) and (iv) uses the k-Means algorithm for grouping images that are semantically similar. The proposed approach has been successfully validated using 33,192 images from the ALOI database, obtaining a mean recall value of 47.4% for searches of images containing objects that are identical to the object query and 20.7% for searches of images containing different objects (albeit visually similar) to the object query.
General TermsContent-based image retrieval, relevance feedback, feature extraction.
KeywordsObject-based image retrieval, scale invariant feature transform, principal component analysis, vector of locally aggregated descriptors, clustering algorithms.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.