This is the slides for the seminar on skin tone based steganography. By GIRISH RAM M, TKM COLLEGE OF ENGINEERING KOLLAM

1. Skin Tone Based Image
Steganography
Girish Ram M
R7121

2. What is Steganography??
Steganography is the ART OF HIDING the secret and
confidentional data in another transmission medium
to achieve secret communication.
Hiding of data in Image is called
Image Steganography.

3. Why do we need Steganography??
Internet serves as an important role for data transmission and
sharing.
It is a worldwide and publicized medium, some confidential
data might be stolen, copied, modified, or destroyed
Security problems become an essential issue.
Encryption is a well-know procedure for secured data
transmission

4. Why not Encryption??
Encryption is the process of encoding messages (or
information) into a form that cannot be easily understood by
unauthorized people.
Disadv of encryption: They make the secret messages
unreadable and unnatural, These unnatural messages usually
attract some unintended observer’s attention.
Therefore, Digital Steganography is a new
security approach for the transmission of confidential data

5. Encryption
Steganography
Original image
Secret data
Stegano image

6. Comparison b/w Steganography &
Watermarking
Both are branches of data hiding.
Watermarking is the technique of embedding digital
marks inside a container(image) so that there is a logical
way of extracting the data embedded.
Here no harming of container.
Steganography uses the cover file to deliver its secret
message while watermarking considers the cover file as
the important data that is to be preserved.
Steganography is used for secret communication while
watermarking is used for authentication.

7. Fig1. Trade off b/w embedding capacity, Robustness and undetectability
in Data hiding
Embedding Capacity
Naive steganographic
technique
Secure steganographic
technique
Digital Watermarking
Undetectability Robustness
Steganography pay attention to the degree of invisibility
while Watermarking pay most of its attribute to the
robustness of the message and its ability to withstand
attacks of removal.

8. Existing Methods for Image
Steganography
LSB(Least Significant Bit) Substitution based.
Transform domain based.
Adaptive method.
Proposed Method
Biometrics based steganography using DWT(Discrete
Wavelet Transform)
Biometric feature used is Skin tone region of images.
In this method secret data is embedded within skin region.
Skin region are not sensitive to HVS(Human Visual System).

9. Overview of method
First: Skin tone detection is performed on input image
using HSV (Hue, saturation, value).
Second: Cover image is transformed in frequency domain
using Haar-DWT.
Payload number calculated.
Finally Secret message is embedded in one of the high
frequency sub-band by tracing skin pixels in that band.
Embedding steps are applied to 2 cases:
With Cropping
Without Cropping
Both cases uses different embedding algorithm.

10. With Cropping
First cropping on input image is performed and only in
cropped region data hiding is performed.
Cropped region works as a key at decoding side.
More Secure.
Without Cropping
Data is embedded in whole Skin region image.
Embedding algorithm attempts to preserve histogram of
DWT coefficients after embedding.

11. Skin Tone Detection Using
HSV
Aim: Discriminate b/w skin & non-skin pixels.
Transforms a given pixel into an appropriate color space.
Uses a skin classifier to label the pixel whether it is a skin
or a non-skin pixel.
Skin detection algo produces a mask – black and white.
White pixel = 1 Skin pixel.
Black pixel = 0  Non-Skin pixel.

12. How to decide weather a pixel is skin or
not skin?
RGB matrix of the given color image is converted into
different color spaces to yield distinguishable regions of
skin or near skin tone.
Mainly two kinds of color spaces are available HSV (Hue,
Saturation and Value) and YCbCr spaces.
It is experimentally found that the distribution of human
skin color constantly resides in a certain range within the
color space.
Here HSV color space is chosen.

13. Cont’d..
RGB image is converted into HSV color space.
In HSV, responsible values for skin detection are Hue &
Saturation.
Extract Hue & Saturatn dimensions into separate new variables
(H & S).
For Skin detection threshold is chosen as
[H1, S1] & [H2,S2].
Sobottaka & Pitas defined a face localization based on HSV.
They found that the threshold range of human flesh as:
Smin= 0.23, Smax =0.68, Hmin =0° & Hmax=50°.

14. HSV Color model
Pure HueWhite Tints
 Hue at 0o is red
 Hue at 120o is green
 Hue at 240o is blue
YCbCr Model
Black
Shade
Tones

15. Converted to HSVOriginal Image

16. DWT and Haar-DWT
This is one of the Frequency domain in which
Steganography can be implemented.
In this work DWT better than DCT.
Haar-DWT is simplest form of DWT.
DWT splits component into numerous frequency sub
bands as:
LL – Horizontally and vertically low pass
LH – Horizontally low pass and vertically high pass
HL - Horizontally high pass and vertically low pass
HH - Horizontally and vertically high pass
Human eyes are more sensitive to the low frequency part
(LL sub-band).
Hence secret message in hidden other 3 parts.

17. Haar-DWT
 Adv: High Computation Speed, Simplicity.
 Performs row and column transformation of Image matrix.
Procedure
1.Haar DWT of an array
2.Using Matrix Multiplication:
 If A is matrix on which we want to perform Haar-DWT, row
transform is performed using
Q=A×P , where P=
 O/P of row transform becomes I/P at column transform.
 Column transformation is obtained using
R= P’× Q where P’ is transpose of P.
 Matrix P performs averaging and differencing operation.
 Result is four sub-bands.
1 0 1 0
1 0 -1 0
0 1 0 1
0 1 0 -1

18. Encoding
Suppose C is original 24-bit color cover image of M×N
size.
C= {xij, yij, zij |1<=i<=M, 1<= j . N, xij, yij, zij in {0, 1,..,
255}}
Let S is secret data. Here secret data considered is binary
image of size a×b.
Two cases of Encoding process:
 With cropping
 Without cropping

19. With Cropping
Let size of cropped image is Mc×Nc where Mc≤M and
Nc≤N and Mc=Nc. i.e. Cropped region must be exact square
as we have to apply DWT later.
Step1:Apply skin tone detection on cover image. This will
produce mask image that contains skin and non skin pixels.
Step2: Cropping is done interactively on image (Mc×Nc).
Note: Cropped area must be an exact square and
cropped area should contain skin region such as face,
hand etc.
 Cropped rectangle will act as key at receiving side.

20. Step3: Apply DWT to cropped area.
This yields 4 sub-bands denoted as HLL, HHL,HLH,HHH
Payload of image is determined based on nos of skin
pixels present in one of high frequency sub-band.
Embedding in LL sub-band affects image quality
greatly.
Step4: Embedding in high frequency HH sub-band is done
only in skin pixels. Skin pixels are traced using skin mask
detected earlier secret data is embedded
Embedding is performed in G-plane and B-plane of RBG
but strictly not in R-plane as contribution of R plane in
skin color is more.
Embedding is done as per raster-scan order.

21. Step5: Perform IDWT to combine 4 sub-bands.
Step6: A cropped stego image of size Mc×Nc is obtained.
Merge the cropped stego image with original image to get
the stego image of size M×N.
For merging coefficients of first and last pixels of cropped
area in original image is required so that R is calculated.
Finally, Stego image is ready for secret communications.

22. Flowchart of With Cropping case of
Embedding Process

24. Without Cropping
There is a major difference is actual embedding algorithm.
In this embedding algorithm, Data is hidden such that the
histogram of the cover image shouldn’t get modified.
Step1:Apply skin tone detection.
Step2: Separate R, G, B planes, Apply DWT to B plane.
Skin pixels from HH sub-band are retrieved and stored in
one matrix.

25. Step3: Apply embedding algorithm to retrieved matrix.
Here, start from first pixel and using pseudorandom
sequence its corresponding pixel for pair is found.
To create a pseudo random sequence Lehmer’s
Congruential generator is used. It generates non
overlapping random sequence. Lehmer invented the
multiplicative congruential algorithm - which is the basis
for many of the random number generators today.
Lehmer's generators involve 3 integer parameters a, c, and
m, and an initial value x0, called the seed
xk+1 = axk + c mod m
Pairs are formed using this equation.

26. Embedding algorithm
• Input- matrix of only skin pixels, let it is S and secret
• message bits of size M
• Output- Modified matrix of only skin pixels
• Begin
• 1] Select non-overlapping, random pair from S. Let it be p1,
p2.
• If count (message bits)=M then
• goto End
• Else
• goto step 2
• 2] if p1=p2 then
• goto step 1 (choose other pair)

27.  3] if p1‚ p2 then
 goto step 4.
 4] if message bit =0 then
 if p1>p2 then
 swap (p1,p2).
 Choose next message bit.
 Else Choose next message bit.
 goto step 1.
 5] if message bit=1 then
 if p1<p2 then
 swap(p1,p2)
 Choose next message bit.
 else Choose next message bit.
 goto step 1.
 End

28. Step4: This modified matrix of only skin pixels is restored
in HH sub-band.
Step5: Apply IDWT to merge all the sub-bands.
Finally, Stego image is ready.

29. Decoding
Case 1- With Cropping
In this case we must need a value of cropped area to retrieve
data.
Suppose cropped area value is stored in ‘rect’ variable.
‘rect’ will act as a key at decoder side.
By tracing skin pixels in HHH sub-band of DWT secret data
is retrieved.

30. Case 2- Without Cropping
In this case extraction of secret data is done without cropping.
Decoding requires finding skin pixel pairs that are used for
embedding.
These pixel pairs are found by generating non-overlapping,
random sequence using Lehmer’s Congruential generator.
 Once correct pixel pairs are found, based on their values
either one or zero of secret data is decided.

31. Flowchart of Decoding Process

32. Performance Measure
Peak signal to noise ratio (PSNR) is used to evaluate
quality of stego image after embedding the secret message.
PSNR is defined as:
Xij and Yij represents pixel values of original cover image
and stego image respectively.

33. PSNR usually adopts dB value for quality judgement.
The larger PSNR is, higher the image quality.
On the contrary smaller dB value means there is a more
distortion.
PSNR values falling below 30dB indicate fairly a low
quality. However, high quality strives for 40dB or more.

34. Capacity and PSNR of 4 final stego
images in proposed method.
CASE A- WITHOUT CROPPING
CASE B- WITH CROPPING

35. Conclusion
Digital Steganography is a fascinating scientific area which
falls under the umbrella of security systems.
Skin Tone based Steganography is presented that uses skin
region of images in DWT domain for embedding secret data.
Two cases for data embedding are considered, with cropping
and without cropping.
According to performance measure results, proposed
approach provides fine image quality. Since PSNR of all
images are above 40db.

36. thank you..
Questions?