Vision Research Lab - Image Adaptive Data Hiding

People

K. Solanki, A. Sarkar, N. Jacobsen, U. Madhow, B.S. Manjunath, S. Chandrasekaran

Objective

The goal of data hiding is embed as much information as possible into a host without causing any perceptual distortion, and still be robust enough to survive benign or malicious attacks.

In other words, there are three conflicting requirements for a data hiding scheme:

Transparency: The host signal should not undergo any perceptual degradation.
Robustness: The embedded signal must survive benign and malicious attacks.
Capacity: Embed as much information as possible into the host.

Our objective is to hide large volumes of data in images, in a manner that causes minimal perceptual distortion, and is robust to "benign" JPEG compression as well as more malicious additive white Gaussian noise (AWGN) attacks.

Demo

» Data Hiding Demonstration

Image-Adaptive Data Hiding

We have designed a high-volume data hiding scheme that uses image-adaptive criteria and powerful error correcting codes, and is robust against a variety of attacks such as JPEG compression, AWGN, wavelet-based compression (JPEG 2000), image resizing, and image tampering. It has the following main ingredients.

Transform domain hiding: We embed data in the DCT transform domain. The possible candidates for embedding are the transform coefficients in the low and mid frequency bands.
Image-adaptive criteria: A novel feature of our method is that from the set of candidate transform coefficients, the encoder employs a local criterion to select a subset of coefficients to embed data. In practice, the use of a local criterion for deciding where to hide is crucial to maintain the image quality under high volume embedding. Two such techniques are proposed which are called the Entropy Thresholding (ET) scheme, and the Selectively Embedding in Coefficients (SEC) scheme.
Coding framework: The use of powerful error correcting codes allows us to deal with desynchronization arising due to locally adaptive hiding. It also provides robustness against a variety of attacks.

Examples

Selective Embedding in Coefficients (SEC) Scheme

Example 1: Lena
Example 2: Peppers
Example 3: Man

Entropy Thresholding (ET) Scheme

Three test images are presented here with data hidden at design quality factors (QF) of 75, 50, and 25 respectively along with number of bits embedded. It should be noted that lower design quality factors indicate more robustness against attacks.

Example 1: Lena
Example 2: Peppers
Example 3: Man

» Robustness against attacks

Tamper Detection and Localization

This coding framework can also deal with image tampering wherein a part of image is replaced by some other image data. Such a tampering can be local or global. In order to survive tampering, the code rate used is further lowered so that we can deal with the errors caused due to the replacement of the image data. Once the hidden bitstream is decoded, localization of the tampered area can be done by finding the locations in the host image where errors have occurred. If the host image has undergone tampering, then most of the errors would be concentrated at the locations where the tampering was done. Such an ability to robustly decode the embedded bitstream and then localize the tampered area can be useful in medical or forensic applications to detect whether a malicious attacker has tampered with the "evidence".

Examples

Tamper Detection and Localization

Acknowledgements

This research is supported by a grant from ONR #N00014-01-1-0380. Program manager: Dr. Ralph Wachter

Publications

2007

A. Sarkar, U. Madhow, S. Chandrasekaran and B. S. Manjunath,
"Adaptive MPEG-2 Video Data Hiding Scheme"
Proc. SPIE Security, Steganography, and Watermarking of Multimedia Contents IX, Jan. 2007.
[abstract] [PDF] [BibTex]
Abstract preview: "We have investigated adaptive mechanisms for high-volume transform-domain data hiding in MPEG-2 video which can be tuned to sustain varying levels of compression attacks. The data is hidden in the unc..." [more]

2004

K. Solanki, N. Jacobsen, U. Madhow, B. S. Manjunath and S. Chandrasekaran,
"Robust Image-Adaptive Data Hiding Based on Erasure and Error Correction"
IEEE Transactions on Image Processing, vol. 13, no. 12, pp. 1627-1639, Dec. 2004.
VRL ID: 141, [abstract] [PDF] [BibTex]
Abstract preview: "Information-theoretic analyses for data hiding prescribe embedding the hidden data in the choice of quantizer for the host data. In this paper, we propose practical realizations of this prescription f..." [more]

2003

K. Solanki, O. Dabeer, U. Madhow, B. S. Manjunath and Shiv Chandrasekaran,
"Robust Image-Adaptive Data Hiding: Modeling, Source Coding, and Channel Coding"
41st Allerton Conference on Communications, Control, and Computing, Oct. 2003.
VRL ID: 125, [abstract] [PDF] [BibTex]
Abstract preview: "This paper provides a summary of our work over the past two years on robust, high-volume data hiding in images. We first present a basic framework for image-adaptive hiding, which allows selection of ..." [more]

2002

N. Jacobsen, K. Solanki, U. Madhow, B. S. Manjunath and S. Chandrasekaran,
"Image-adaptive hidgh-volume data hiding based on scalar quantization"
Proc. IEEE Military Communications Conference (MILCOM), Anaheim, CA, USA, vol. 1, pp. 411-415, Oct. 2002.
VRL ID: 110, [abstract] [PDF] [BibTex]
Abstract preview: "Information-theoretic analyses for data hiding prescribe embedding the hidden data in the choice of quantizer for the host data. In this paper, we consider a suboptimal implementation of this presc..." [more]
K. Solanki, N. Jacobsen, S. Chandrasekaran, U. Madhow, and B. S. Manjunath,
"High-volume data hiding in images: Introducing perceptual criteria into quantization based embedding"
Proc. ICASSP, vol. 4, pp. 3485-3488, May. 2002.
VRL ID: 96, [abstract] [PDF] [BibTex]
Abstract preview: "Information-theoretic analyses for data hiding prescribe em-bedding the hidden data in the choice of quantizer for the host data. In this paper, we consider a suboptimal implementation of this prescri..." [more]