Efficient image copy detection using multi-scale fingerprints
Files
Date
Supervisor
Item type
Journal Article
Degree name
Journal Title
Journal ISSN
Volume Title
Publisher
IEEE
Abstract
Inspired by multi-resolution histogram, we propose a multi-scale SIFT descriptor to improve the discriminability. A series of SIFT descriptions with different scale are first acquired by varying the actual size of each spatial bin. Then principle component analysis (PCA) is employed to reduce them to low dimensional vectors, which are further combined into one 128-dimension multi-scale SIFT description. Next, an entropy maximization based binarization is employed to encode the descriptions into binary codes called fingerprints for indexing the local features. Furthermore, an efficient search architecture consisting of lookup tables and inverted image ID list is designed to improve the query speed. Since the fingerprint building is of low-complexity, this method is very efficient and scalable to very large databases. In addition, the multi-scale fingerprints are very discriminative such that the copies can be effectively distinguished from similar objects, which leads to an improved performance in the detection of copies. The experimental evaluation shows that our approach outperforms the state of the art methods.Description
Inspired by multi-resolution histogram, we propose
a multi-scale SIFT descriptor to improve the discriminability.
A series of SIFT descriptions with different scale are first
acquired by varying the actual size of each spatial bin. Then
principle component analysis (PCA) is employed to reduce them
to low dimensional vectors, which are further combined into one
128-dimension multi-scale SIFT description. Next, an entropy
maximization based binarization is employed to encode the
descriptions into binary codes called fingerprints for indexing
the local features. Furthermore, an efficient search architecture
consisting of lookup tables and inverted image ID list is designed
to improve the query speed. Since the fingerprint building is
of low-complexity, this method is very efficient and scalable to
very large databases. In addition, the multi-scale fingerprints
are very discriminative such that the copies can be effectively
distinguished from similar objects, which leads to an improved
performance in the detection of copies. The experimental evaluation shows that our approach outperforms the state of the art
methods.
Source
IEEE MultiMedia Magazine. January-March 2012 (vol. 19 no. 1)
pp. 60-69.
Publisher's version
Rights statement
© 2011 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.