Secure image retrieval through feature protection

Lu, Wei; Varna, Avinash L.; Swaminathan, Ashwin; Wu, Min

doi:10.1109/icassp.2009.4959888

Cited by 115 publications

(84 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…For such cases, the integer vectors can be naively mapped into binary feature vectors which are elements of a real-valued ℓ 2 metric space, such that the squared ℓ 2 distance between the binary vectors is equal to the ℓ 1 distance between the original integer feature vectors. 17,39 In particular, without loss of generality, consider that each element u i , i ∈ {1, 2, . .…”

Section: Extension To Nn Search Based On Non-euclidean Distance Measuresmentioning

confidence: 99%

Quantized embeddings: an efficient and universal nearest neighbor method for cloud-based image retrieval

2013

View full text Add to dashboard Cite

We propose a rate-efficient, feature-agnostic approach for encoding image features for cloudbased nearest neighbor search. We extract quantized random projections of the image features under consideration, transmit these to the cloud server, and perform matching in the space of the quantized projections. The advantage of this approach is that, once the underlying feature extraction algorithm is chosen for maximum discriminability and retrieval performance (e.g., SIFT, or eigen-features), the random projections guarantee a rate-efficient representation and fast server-based matching with negligible loss in accuracy. Using the Johnson-Lindenstrauss Lemma, we show that pair-wise distances between the underlying feature vectors are preserved in the corresponding quantized embeddings. We report experimental results of image retrieval on two image databases with different feature spaces; one using SIFT features and one using face features extracted using a variant of the Viola-Jones face recognition algorithm. For both feature spaces, quantized embeddings enable accurate image retrieval SPIE Conference on Applications of Digital Image Processing XXXVIThis work may not be copied or reproduced in whole or in part for any commercial purpose. Permission to copy in whole or in part without payment of fee is granted for nonprofit educational and research purposes provided that all such whole or partial copies include the following: a notice that such copying is by permission of Mitsubishi Electric Research Laboratories, Inc.; an acknowledgment of the authors and individual contributions to the work; and all applicable portions of the copyright notice. Copying, reproduction, or republishing for any other purpose shall require a license with payment of fee to Mitsubishi Electric Research Laboratories, Inc. All rights reserved. ABSTRACTWe propose a rate-efficient, feature-agnostic approach for encoding image features for cloud-based nearest neighbor search. We extract quantized random projections of the image features under consideration, transmit these to the cloud server, and perform matching in the space of the quantized projections. The advantage of this approach is that, once the underlying feature extraction algorithm is chosen for maximum discriminability and retrieval performance (e.g., SIFT, or eigen-features), the random projections guarantee a rate-efficient representation and fast server-based matching with negligible loss in accuracy. Using the Johnson-Lindenstrauss Lemma, we show that pair-wise distances between the underlying feature vectors are preserved in the corresponding quantized embeddings. We report experimental results of image retrieval on two image databases with different feature spaces; one using SIFT features and one using face features extracted using a variant of the Viola-Jones face recognition algorithm. For both feature spaces, quantized embeddings enable accurate image retrieval combined with improved bit-rate efficiency and speed of matching, when compared with the underlying featur...

show abstract

Section: Extension To Nn Search Based On Non-euclidean Distance Measuresmentioning

confidence: 99%

Quantized embeddings: an efficient and universal nearest neighbor method for cloud-based image retrieval

2013

View full text Add to dashboard Cite

show abstract

“…The dataset consists of 1000 images grouped in 10 classes; each class contains 100 color images [14]. These classes are categorized as Africa, Beach, Buildings, Buses, Dinosaurs, Flowers, Elephants, Horses, Mountains, and Food.…”

Section: Resultsmentioning

confidence: 99%

Encrypted Image Retrieval System based on Features Analysis

Gaata

Hantoosh²

2018

MJS

View full text Add to dashboard Cite

Content-based search provides an important tool for users to consume the ever-growing digital media repositories. However, since communication between digital products takes place in a public network, the necessity of security for digital images becomes vital. Hence, the design of secure content-based image retrieval system is becoming an increasingly demanding task as never before. In this paper, the secure CBIR with additional improvement for the image retrieval has been presented. The proposed system consists of six phases briefly described as follows: first, feature extraction phase, which produces the low-level quantitative description of the image (color and texture) that used in calculation of similarity score and image indexing. Second, indexing for search process phase, hash table and bloom filter were employed for classification. Third, feature encryption phase, where content protection is performed using a method developed by us (including Chaotic Logistic Map). Fourth, image encryption phase, the chaos and stream cipher systems were applied as an image encryption system in order to achieve image security. Fifth, the retrieval phase, which provides a group of images replying the query based on the similarity score between images, calculated using the extracted features from each image. Finally, Relevance feedback phase, a technique that attempts to capture the user's needs through iterative feedback. Although the system proved its efficiency in search performance (with 88% of average precision), security strength, and computational complexity, it does not mean the optimal system is designed, since some weakness points still can be found that are suggested to be improved as a future work.

show abstract

“…Both image operations and cryptographic primitives are combinely studied in [12], [13]. In [4], [14], the authors study about the oblivious/private retrieval of images. In [4], the authors focus on oblivious comparison of image features and claim that this step alone could suffice for obliviously retrieving the images from the database.…”

Section: A Limitation Of Prior Researchmentioning

confidence: 99%

“…In [4], [14], the authors study about the oblivious/private retrieval of images. In [4], the authors focus on oblivious comparison of image features and claim that this step alone could suffice for obliviously retrieving the images from the database. However, the features considered in their implementation are color histograms.…”

Section: A Limitation Of Prior Researchmentioning

confidence: 99%

See 1 more Smart Citation

An Oblivious Image Retrieval Protocol

Sabbu

Umadevi

Srinathan

et al. 2011

2011 IEEE Workshops of International Conference on Advanced Information Networking and Applications

View full text Add to dashboard Cite

Abstract-We consider the novel problem of oblivious retrieval of images from an outsourced database server. Due to the complexity of managing images, in modern enterprise networks, individual users and organizations are using the services of external servers to store the images. The problem is to query and retrieve images while simultaneously satisfying the criteria of both the user and the database privacy. Oblivious transfer protocols are capable of ensuring the privacy of both the user and the server privacy at the same time. But in their native form, oblivious transfer protocols may not be suitable for querying and retrieving images due to the large size of the images. In this paper, we describe the first such attempt to develop an oblivious retrieval protocol that can achieve the goals of user and database privacy. which allows the user to query for the images of interest. In this phase, the database server does not know the exact user query and the user does not learn of any other images in the database. In the next phase, the user deploys an oblivious transfer protocol to retrieve the necessary decryption for the selected images without revealing the actual set of decryption keys selected. We have evaluated our protocol using an experimental setting and show that the results are a reasonable trade-off when compared to the security requirements being satisfied. We have implemented a working prototype and tested the protocol on a database of 9000 large real-life images.

show abstract

Secure image retrieval through feature protection

Cited by 115 publications

References 7 publications

Quantized embeddings: an efficient and universal nearest neighbor method for cloud-based image retrieval

Quantized embeddings: an efficient and universal nearest neighbor method for cloud-based image retrieval

Encrypted Image Retrieval System based on Features Analysis

An Oblivious Image Retrieval Protocol

Contact Info

Product

Resources

About