13-11-2012, 04:20 PM
Steganographic Embedding in JPEG Images with Visual Criterion
Steganographic Embedding.pdf (Size: 215.56 KB / Downloads: 26)
Abstract
In this paper, we present a new information hiding
scheme in JPEG images to achieve a good embedding efficiency
considering visual criterion. We construct an embedding impact
model based on human visual system, and then assign each cover
element a flipping cost which would be the key parameter during
the embedding procedure. In this way, the proposed method can
minimize the total embedding impact via Viterbi algorithm,
meanwhile improve the visual quality of the stego medium. The
experimental results later show that the proposed information
hiding system can perform well in different types of images.
INTRODUCTION
Recent years, research on data hiding is becoming
important for protecting the confidential information. In data
hiding, especially steganography, the secret messages are
embedded imperceptibly into cover files (such as digital
images, audios, and videos) by slightly modifying some of the
cover elements (pixels, DCT coefficients, etc.). The media
with the embedded secret known as the stego medium must
appear to be similar to its original so that the stego medium
cannot arouse suspicion. However, the anti-data hiding
research has also rapidly developed to detect the presence of
secret messages based on revealing visual or statistical
abnormalities in the stego medium. Generally speaking, the
more the embedded data, the more vulnerable the system will
be to the anti-data hiding attempts.
HVS Embedding Impact Model
This section gives human visual embedding impact
model based on the concept of just noticeable differences
(JND) [7], which can capture both threshold effects and
spatial-frequency sensitivity of the HVS. Fusions of
luminance masking, contrast masking and pooling, JND can
reflect the largest modification value under the unawareness.
CONCLUSION
The proposed method achieves a good rate-embedding
efficiency performance as well as guarantees the high visual
quality. Introduction of the Watson’s work, we build a HVS
embedding distortion model which assigns an imbedding cost
for each DCT coefficient of the cover elements. The
syndrome-trellis codes help to minimize the total imbedding
impact. Our experimental results indicate that it outperforms
the F5 significantly in visual quality and statistic
characteristics under different types of images and different
payload.