Premium
Secure content‐based image retrieval using modified Euclidean distance for encrypted features
Author(s) -
Majhi Mukul,
Pal Arup Kumar,
Islam SK Hafizul,
Khurram Khan Muhammad
Publication year - 2021
Publication title -
transactions on emerging telecommunications technologies
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.366
H-Index - 47
ISSN - 2161-3915
DOI - 10.1002/ett.4013
Subject(s) - chrominance , computer science , image retrieval , encryption , histogram , pattern recognition (psychology) , discriminative model , artificial intelligence , feature (linguistics) , visual word , euclidean distance , feature vector , content based image retrieval , image (mathematics) , luminance , linguistics , philosophy , operating system
Abstract In this article, a secure image retrieval scheme is proposed, which focuses on providing satisfactory retrieval results, and the framework searches relevant images even in an encrypted domain without compromising the performance of the retrieval process. Initially, bit‐level features have been endeavored from the luminance component of the image, from which statistical parameters are computed to generate more intrinsic values. These values are subsequently divided into bins to configure two histograms, which effectively reduce the length of the feature vector. These histograms are then eventually combined with quantized chrominance features to enhance the discriminative property of the feature vector. Since the proposed scheme is in the encrypted domain, conventional similarity measure distance for the image is not well suited. So, a modified Euclidean distance is incorporated, which is modeled to work with encrypted features. To comprehend the security, a piecewise logistic map sequence is considered, where seed values are assimilated to generate two secret keys. As a result, not only the system provides an efficient, secure retrieval system but also cryptographic components have no impact on its retrieval efficiency, and satisfactory results are obtained. Experimental results on Corel‐1K and GHIM‐10K illustrate decent performance in retrieval as compared to existing work in the retrieval domain.