A Source and Channel Coding Approach to Data ... - Semantic Scholar

A Source and Channel Coding Approach to Data Hiding with Application to Hiding Speech in Video* Debargha Mukherjee, Jong Jin Chae, and Sanjit K. Mitra Department of Electrical and Computer Engineering, University of California, Santa Barbara, CA 93106. Email: (debu, chaejj, mitra)@iplab.ece.ucsb.edu Abstract: Digital data hiding is a technology being developed for multimedia services, where non-trivial amounts of signature data is invisibly hidden inside a host data source by the owner before the latter is freely distributed. Only those authorized can recover the hidden data from the host, even after the latter has undergone standard transformations such as compression. In this work we adopt a quantitative source and channel coding approach to hiding large amounts of compressible signature data inside the raw host. The signature data is source coded by vector quantization, and the indices are embedded in the host by perturbing it using orthogonal transform domain vector perturbations. The transform coefficients of the parent data are grouped into vectors, and the vectors are perturbed using noise-resilient channel codes derived from multidimensional lattices. The perturbations are constrained by a maximum allowable mean-squared error that can be introduced in the host. The generic approach is readily adapted to make retrieval possible even for applications where the original host is not available to the retriever. In this work, the scheme is applied to hiding speech in video. The host video is wavelet transformed frame by frame, and vectors of coefficients are perturbed using lattice channel codes to represent hidden vector quantized speech. The embedded video is subjected to H.263 compression before retrieving the hidden speech from it. The retrieved speech is intelligible even with large compression ratios of the host video.

Figure 1 presents a common schematic of the data hiding and watermarking problem. The original host is modified using the signature data in a deterministic fashion before distribution. As a result of embedding, a mean-squared-error MSEH is introduced into the embedded host. To ensure transparency of embedding, the MSEH value should be below a certain desired level. While in watermarking, the allowable MSEH is very small, and so is the amount of signature data, in data hiding, the focus is more on hiding larger amounts of signature data at the expense of a higher allowable MSEH. On distribution, the host typically undergoes compression and other standard transformations. The extraction process may or may not, depending on the nature of the application, require knowledge of the original host, to estimate the hidden signature from the ‘noisy’ embedded host that is received. After extraction, it is desired that the channel mean-squared-error MSES between the original signature and the extracted signature be as low as possible. MSEH

Original Host

Embedding

Embedded Host

Distribution

Compression/ Other Transformations

Signature MSES

1. Introduction With the rapid growth in the mass of multimedia data freely available to the users of the Internet, a mechanism for data security or copyright protection presents an overwhelming challenge. Digital watermarking [1]-[9] is an emerging technology, where small amounts of a signature information is invisibly hidden in a host data source by the owner before the host is distributed freely. The challenge is to enable the owner to retrieve his original signature from the distributed image to check authenticity, even after it has undergone significant transformations such as compression. A closely related problem to digital watermarking is data hiding. Here the focus is on hiding larger amounts of data in a host without making it obvious, so that only those authorized with a certain esoteric knowledge of ‘how to’ can retrieve it. Data hiding has applications in secure communications where an insecure but readily available medium such as the Internet is used to transmit hidden data. It can also be used for transmitting different kinds of information securely over an existing channel dedicated for transmitting something else, such as transmitting hidden speech over a channel meant for transmitting H.263 video, as in this work. Since a substantial amount has already been invested in the development of the software and hardware infrastructure for standard-based data transmission, it makes monetary sense to try to use the same for transmission of secure or non-standard data.

* This work was supported by ONR Grant N00014-95-1-1214.

Noisy Embedded Host Extraction

Extracted Signature

Original Host

Figure 1. Data Hiding

In this work, we show that the above dual problems of data hiding and watermarking, readily map to the source and channel coding problem in digital communications. As such, established concepts from digital communications could be used to solve this problem. In Section 2 we discuss in detail our data hiding approach where vectors of orthogonal transform coefficients of the host are perturbed using lattice channel codes. In Section 3, we present an example application where vector quantized speech is hidden in QCIF video.

2. Data Hiding using Vector Perturbations The host data is orthogonally transformed before embedding the hidden signature in it. The transform is not essential because a raw image or video is by itself an expansion on the standard bases. However, it may lead to some advantages. Let us consider a host data source ( X 1, X 2, …, X N ) transformed orthogonally to a set of N coefficients ( C 1, C 2, …, C N ) . The transform-domain embedding process perturbs the coefficients into a new set of coefficients given by ( Cˆ 1, Cˆ 2, …, Cˆ N ) . The inverse transformation then yields the embedded host ( Xˆ 1, Xˆ 2, …, Xˆ N ) . Since the transformation is orthog-

k-dim space

Signature

Coding

sequence of symbols from

k-dim space si

s3 Embedding

Vj

s2

a Q-ary alphabet {s1,s2,...,sQ}

Noisy

kP

c

Channel

Extracted Signature

Decoding

sequence of extracted symbols

Wj Vˆ j si

s3 Symbol Extraction

s2

Channel

Vj

sQ s1

Noisy

sQ s1

Decision Boundaries

Figure 3. The Extraction Principle

Figure 2. The Embedding principle

onal, the mean-squared-error introduced in the coefficients is exactly equal to the mean-squared-error introduced in the host data. That is, N

MSE H

1 = ---- ⋅ N

∑

N 2 1 X i – Xˆ i = ---- ⋅ N

i=1

∑

ˆ 2 Ci – C i

(1)

i=1

Now, a transparency constraint is imposed on the value of MSEH. This specifies a maximum value P which upper bounds MSEH for a given application: N

1 ---- ⋅ N

∑ i=1

N

2 Xi – Xˆ i < P

1 ⇒ ---- ⋅ N

∑

ˆ 2