Abstract:
Image retrieval systems help users to browse and search among extensive images in
real-time. Various image retrieval techniques have been developed over the years. Content
based image retrieval is a technique employed to search and retrieve similar images based
on their visual content (query image) rather than text-based query. On the other hand,
secure content-based image retrieval is meant to retrieve similar images from encrypted
images. At present, various secure content-based image retrieval schemes have been
developed which encrypt images first and extract features subsequently. However, none of
existing schemes uses NIST approved / globally accepted encryption nor there are any
techniques available which gives liberty to user to use any encryption scheme resulting in
restricting user’s liberty, which is contradictory to multi-user environment of cloud. In this
research, we propose a novel paradigm which gives independence to users to use any
encryption primitive as per their security needs. Applicable to optimized JPEG images, our
proposed system employs segmented encryption to encrypt image data segment of JPEG
images. Subsequently, huffman tables; extracted from encrypted images are used for
retrieval tasks employing machine learning in both supervised and unsupervised domain.
Experiments reveal that our scheme achieves excellent performance in terms of efficiency.
