29-03-2011, 04:01 PM
SUBMITTED BY
I.NIJANTHAN
[attachment=11271]
ABSTRACT:
Information hiding in JPEG2000 compressed images is investigated in this paper. Our paper mainly concern with the challenges of covert communication in this state-of-the-art image codec are analyzed and a steganographic scheme is then proposed to reliably embed high-volume data into the JPEG2000 bit-stream. A special mode of JPEG2000 is employed, and its usage and functions are explained and justified.
JPEG2000 is an upcoming still image coding standard. This new standard complements JPEG by providing several important features such as resolution/quality progressive image transmission, better resilience to bit-errors, and Region of Interest (ROI) coding, etc. It is believed that JPEG2000 will be used widely and its rich features will benefit many emerging applications. As many images will be compressed by JPEG2000 in the near future, it is worthwhile to investigate how to hide high-volume data in JPEG2000 compressed images efficiently. This is the main objective of the paper.
II. REVIEW OF JPEG2000 CODING:
The block diagram of JPEG2O00 is shown in Fig. I. In the encoder side, the original image first undergoes the forward image transform, which includes the inter-component transform and the intra-component transform (i.e. the wavelet transform). The resulting, wavelet coefficients are then quantized and coded. Scalar or Trellis Coded Quantization is used, which may cause some information loss if the image is lossy compressed. The coding paradigm of JPEG2000 can be viewed as a two-tiered structure as shown in Fig. 1 and will be explained in detail below. Rate control is applied in quantization and coding steps to achieve the targeted bit-rate. The decoder side basically reverses the operations by decoding and dequantizing the bit-stream and applying the inverse image transform to reconstruct the image.
Now, let us take a closer look at the two-tiered coding structure in JPEG2000. We illustrate the concept from the encoder part. In tier-1 coding, the quantization indices for each sub band are partitioned into code blocks, which arc independently coded using a bit-plane coder. More specifically, the code block is coded one bit-plane at a time starting from the most significant bit-plane to the least significant bit-plane. Each individual bit-plane is coded with three coding passes. The first coding pass is the significance propagation pass, which conveys significance and necessary sign information for samples that have not yet been found to be significant and are predicted to become significant. The second coding pass is the magnitude refinement pass all bits that became significant in a previous bit-plane are conveyed in this pass by using binary symbols. The final pass is the cleanup pass, in which all bits that have not yet been coded during the previous two passes are encoded. The symbols generated from the significance propagation and the magnitude refinement passes can be either raw coded or entropy coded by a context-based adaptive binary arithmetic coder, i.e. MQ coder.
The cleanup pass is run-length coded and always processed by the MQ coder. The output of the tier-1 encoding process is therefore a collection of compact representations of coding passes for the code blocks. Tier-2 coding operates on the summary information of code blocks, which determines block contributions to the final code stream. The bit-stream of each code-block is truncated in an optimal way so as to minimize distortion subject to the bit-rate constraint. Basically, truncation can only happen at end of a coding pass. Feasible truncation points have been further identified as those located within the convex hull of the rate-distortion curve. To minimize the distortion with the targeted bit-rate constraint, the exact truncation point is chosen from these feasible ones in each block after the statistics of a collection of code blocks are available.
The coding passes are then packaged into packets and output to form the final code stream. The ordering of packets in the code stream facilitates progressive transmission of the image by fidelity, resolution or component. Since the rate distortion algorithm of the tier-2 coding is applied after all sub band samples have been compressed in ticr-1 coding, the rate-control mechanism of JPEG2000 can thus be referred to as Post-Compression Rate Distortion (PCRD) optimization. Besides, it should be noted that some coding passes may be discarded by this optimized truncation procedure so that the ticr-2 coding is another primary source of information loss in the coding path besides quantization.
III.INFORMATION HIDING IN JPEG2000
A. Challenges of Information Hiding
Our objective is to, develop an information hiding scheme under the framework of JPEG2000 so that a high volume of data can be secretly transmitted to the intended recipient in a more reliable fashion. First, we have to determine an appropriate position in JPEG2000 coding flow for information hiding. From the structure given in Fig. 1, there are three positions to be considered. We examine their suitability for information hiding below.
(1) Image Transform
After the intra-component image transform, the image data are transformed to wavelet coefficients. If we modify the data at this stage, the scheme will be equivalent to many existing wavelet-based watermarking algorithms, which may take other wavelet-based codec’s as attacks. For digital watermarking, the payload is usually low, and multiple embedding with the majority detection or the spread-spectrum concept can be applied. The embedded information can thus have sufficient robustness against lossy compression of another codec. However, multiple embedding is not suitable in the data hiding application as the required payload is high and we have to make efficient use of the already limited bandwidth.
(2) Quantization
Quantization is an important step in image compression, which reduces certain visual redundancy for efficient coding. As mentioned in Section II, quantization is the primary source of information loss. We can avoid losing the hidden data due to coarser quantization by embedding them in the quantization indices. This solution works for JPEG (as many JPEG-based information hiding schemes operate on the quantization indices), but is not good for JPEG2000. It should be noted that wavelet-based coders usually truncate the compressed bit-stream to fulfill the targeted bit-rate. In JPEG2000, PCRD optimization strategy is adopted so that the truncation mechanism is activated after the_ whole image has been compressed. If embedding the information at this stage, we cannot predict exactly which quantization index or bit-plane of an index will be included in the final code stream. The embedded information will not be perfectly recovered unless the lossless compression mode is chosen.
(3) Coding
If the information is embedded in the output of tier-2 coding, i.e. the JPEG2000 packets, is can be guaranteed that all the embedded information will be received-without error and in a correct order because we avoid the two major sources of information loss, i.e. quantization and bit-stream truncation. However, we will have difficulty in modifying the packets for information embedding since the bit-streams may have been compactly compressed by the arithmetic coder. Careless modification could result in failure of expanding the compressed image.
