INVESTIGATION OF INHERENT ROBUSTNESS OF PNG IMAGES FOR LSB STEGANOGRAPHY AND ROLE OF BINARY GOLAY CO

1
INVESTIGATION OF INHERENT ROBUSTNESS OF PNG
IMAGES FOR LSB STEGANOGRAPHYANDROLE OF BINARY
GOLAY CODES IN ENHANCING THE ROBUSTNESS

• BY
• KHIZAR HAYAT KHAN
• Supervised By
• Dr. Tasneem Shah

2
Agenda

• Essential Concepts (slides 3-10)
• Methodology (slide 11)
• Embedding Strategies Employed (slide 12)
• Robustness in the Context of Hue Settings
• Robustness in the Context of Saturation Settings
• Robustness in the Context of Lightness Settings
• Robustness in the Context of Contrast Settings
• Robustness in the Context of Gamma Settings
• Robustness in the Context of Blurredness and
  Sharpness.
• Recommendations.

3
Steganography

• literally means, covered writing
• It is the art of concealment of messages inside
  some innocent looking media, e.g. text, audio,
  video, still images etc.
• an old art Greeks and Chinese
• The recent interest in the field may be traced
  back to the Simmons formulation of the famous
  prisoners problem in 1983.
• Alice Bob and Willie

4
General Scheme

5
Classification of Steganographic Techniques

6
Characteristics of Steganographic Techniques

• Four main factors that characterize the data
  hiding techniques in steganography
• Hiding Capacity the size of information that can
  be hidden relative to the size of the cover.
• Perceptual Transparency It is important that the
  embedding occur without significant degradation
  or loss of perceptual quality of the cover.
• Robustness the ability of embedded data to
  remain intact if the stego-image undergoes
  transformations
• Tamper Resistance refers to the difficulty for
  an attacker to alter or forge a message once it
  has been embedded.

7
Embedding Techniques

• Least significant bit (LSB) insertion
• Embed the message bits directly into the
  least-significant bit plane of the cover image in
  a deterministic sequence.
• Public Key Steganography
• Requires the pre-existence of a shared secret
  key to designate pixels which should be tweaked.
  In this case both the sender and the receiver
  must have this secret.
• Transform domain based embedding
• Embed the data by modulating coefficients in a
  transform domain, such as DCT, DFT, DWT. ) hides
  the data within noise which is then added to the
  cover. The noise is of the type usually incurred
  during the image acquisition process. Such a
  noise is imperceptible to humans if kept to
  limited extent.
• Masking and filtering techniques
• Masking refers to the phenomenon where a signal
  can be imperceptible to an observer in the
  presence of another signal (referred to as the
  masker.)
• Embed information to perceptually significant
  areas of the image , i.e. camouflage.

8
Image Formats

• TIFF- Tagged Image File Format (TIFF)
• BMP- This is a system standard graphics file
  format for Microsoft Windows and hence
  proprietary and platform dependent
• GIF The Graphics Interchange Format
• JPEG - A creation of Joint Photographic Expert
  Group
• PNG - Portable Network Graphic

9
Error Correcting Codes (ECC)

• Hamming Codes (n 2r-1, k 2r-1-r, d 3)
  codes
• Reed-Solomon Codes The most widely used
  Reed-Solomon code is RS(255,223) in which each
  codeword contains 255 code word bytes, of which
  223 bytes are data and 32 bytes are parity. For
  this code n 255, k 223 and e 16.
• Golay Codes Golay (23, 12, 7)

10
Binary Golay (23, 12, 7) Code

• would correct all error patterns of Hamming
  weight (number of nonzero bits) up to 3.
• The total number of these error patterns or the
  number of syndromes is 2048 since there are 11
  parity check bits and 211 2048.
• The number of patterns with a particular number
  of errors x (x 3) can be calculated by using
  the formula.

11
Our Methodology

• Synthesize the image I of particular RGB value
• Embed the message M (Golay-encoded or otherwise)
  into I to get the stego image S S I M
• S S j 0 // S for manipulated stego image.
• Select one property from the set V hue,
  saturation, lightness, contrast, gamma,
  blurredness, sharpness and assign its default
  value to P
• While (S - I M)
• P j
• Apply P to S and Save S
• Note j-1
• PP-j j0
• While (S - I M)
• P-- j--
• Apply P to S and Save S
• Note j-1
• Repeat steps 4-9 for all the members of V one by
  one
• Repeat steps 1-10 for all possible RGB values at
  the interval of 16

12
Embedding Strategies Employed

• For our purpose we had developed the following
  programs
• Program R that embedded a bit in the LSB of the R
  part of the pixel only.
• Program G that embedded a bit in the LSB of the G
  part of the pixel only.
• Program B that embedded a bit in the LSB of the B
  part of the pixel only.
• Program RGB that embedded 3 bits per pixel, one
  each in the LSBs of each of the three parts of
  the pixel, i.e. R, G and B.
• Program RGB2 that embedded 6 bits per pixel, 2
  each in the last two LSBs of each of the three
  parts of the pixel, i.e. R, G and B.
• Miscalaneous programs were developed to embed
  bits in the various pair combinations of R, G and
  B

13
Robustness in the Context of Hue Settings

• The Hue of a pixel refers to its basic color -
  red or yellow or violet or magenta, for instance.
  It is usually represented in the range of 0 -
  360, referring to the color's location (in
  degrees) around a circular color palette.
• Other properties of the image were kept constant
  at there defaults only hue settings were changed
  as a function of RGB values.
• The maximum allowable changes (both on left and
  right) in hue settings were noted for each of the
  pixel value possible at the interval of 16.

14
Robustness in the Context of Hue Settings

• Some Details
• When stego image was subjected to changes in hue
  settings it was revealed that pixel values with
  RGB, i.e. gray scale pixels, were the most
  robust.
• The robustness decreased considerably, however,
  with increase in the embedding density. Thus
  lower embedding density showed greater
  robustness.
• ECC had no effect on the robustness to change in
  hue settings.

15
Robustness in the Context of Saturation Settings

• Saturation is the amount or strength of a color
  present in a pixel.
• Now maximum allowable change in saturation
  settings was studied as a function of RGB pixel
  values Reference
• In the majority of cases, right side inherent
  robustness, to change in saturation settings, was
  more than satisfactory.
• Few pixel values showed good results on the left
  side.

16
Robustness in the Context of Saturation Settings

• The behavior of blue part of the pixel was
  peculiar in the sense that the stego image was
  more vulnerable to corruption when data bit was
  embedded in blue LSB. See some examples here
• Increase in embedding density reduced the
  robustness to saturation changes. See
• Positive impact was noted when Golay code was
  employed as ECC. Some Examples.
• Binary Golay Code also covered the disadvantage
  that the blue byte had, to a greater extent. not
  in case of B127.

17
Robustness in the Context of Lightness Settings

• Red and green bytes showed good robustness in
  some cases. Ref
• The blue byte showed similar vulnerability when
  lightness variable was taken into account.
• very few RGB values showed non-zero robustness to
  lightness changes when blue LSB was involved in
  embedding. see
• This blue syndrome made the three bits per
  pixel embedding (one each in the three LSBs)
  suffer because blue byte was involved in
  embedding. c.f.

18
Robustness in the Context of Lightness Settings

• ECC improved the situation only in the case of
  blue LSB embedding, when a single bit was
  embedded per pixel, and brought it at par with
  green/red LSB embedding.
• ECC did not improve situation much with three
  bits per pixel embedding.
• Thats why we changed our strategy and excluded
  the blue byte from embedding for increased
  embedding density.
• This gave good results.

19
Robustness in the Context of Contrast Settings

• Contrast is a method of altering the brightness
  relationship between the higher and lower color
  ranges of an image.
• Enhancement of contrast will make dark areas
  darker and bright areas brighter.
• Robustness to variations in contrast settings
  largely depended on the value of the byte of the
  pixel to which embedding was made. c.f.
• Byte values of 0, 127 and 255 provided the best
  robustness.

20
Robustness in the Context of Contrast Settings

• Where more than one bit was embedded per pixel in
  distinct bytes the byte-value with low individual
  robustness dictated the terms.
• Increase in embedding density proportionally
  decreased robustness, especially for the
  byte-value of 0, 127 and 255. see
• Binary Golay Code had no role in this regard,
  though the recovered message was legible up to
  some extent beyond the max allowable change.

21
Robustness in the Context of Gamma

• Correct gamma value enables a picture to display
  properly on different platform without losing its
  quality during transformation.
• Gamma Correction uses the following function
• Output Image (Input Image)(1/Gamma).
• There were isolated values or ranges rather than
  a continuous range of allowed change in gamma
  around and adjacent to the default value of 1.00.
  
• Like a line or band spectrum some changes were
  allowed and some not.
• As we moved to extreme R-values from R127 three
  trends were noticed, viz.
• Coagulation, i.e. tendency of neighboring
  points to merge into ranges and expansion in the
  length of range.
• Increase in population of working points/ranges
• Tendency of the working points to shift to the
  middle, i.e. concentrate around the gamma value
  of 1.00.

22
Robustness in the Context of Gamma Settings

• The byte to which embedding was made, contributed
  more to robustness when its value was close to
  the extreme values of 0 and 255. ref
• Blue LSB embedding was again least robust as
  compared to its two other counterparts.
• Robustness was markedly reduced with the increase
  in embedding density. ref
• ECC again improved the results, which were
  otherwise poorer when blue byte was involved in
  embedding.

23
Robustness in the Context of Blurredness and
Sharpness

• Blurring makes an image turbid-looking, i.e.
  visibility of the pixels to the viewer become
  affected out of focus.
• To a certain extent, Sharpening of image is
  possible, although it's somewhat of a trick and
  only works up to a certain point, and results can
  often include undesirable artifacts.
• The sharpen operator actually works by increasing
  the contrast between areas of transition in an
  image, which the eye perceives as sharpness.
• Lost information can never be created through
  sharpness operation.

24
Robustness in the Context of Blurredness and
Sharpness

• The greatest role played by binary Golay Code was
  when blurredness or sharpness filter was applied
  to mutilate the stego image.
• There was originally zero robustness to either
  side change in blurredness or sharpness.
• For low embedding density there was almost 100
  robustness on the left when ECC was involved. On
  positive side too the results improved a lot.

25
Robustness in the Context of Blurredness and
Sharpness

• For higher embedding density of six bits per
  pixel there was no robustness on any either side
  change in sharpness and positive side change in
  blurredness upon the involvement of binary Golay
  Code.
• There was, however, 100 robustness to any change
  in blurredness on the negative side.

26
Recommendations

• Based on the results above we propose three main
  points
• Since the blue axis of RGB plane is the most
  vulnerable to a change in image settings, it must
  be used for embedding only after careful
  consideration.
• One should avoid embedding too many bits per
  pixel unless otherwise absolutely necessary high
  embedding density (bpp) leads to low robustness.
  There would be, of course, low perceptual
  transparency.

27
Recommendations (Contd.)

• A convenient ECC must be employed for robust
  steganographic embedding as it improved things in
  two ways.
• Firstly, ECC improved the robustness across the
  board for all the three axes of the RGB plane,
  especially against blurredness and sharpness
  changes.
• Secondly, it made up for the disadvantage that
  blue axis of the RGB plane had in relation to
  other axes, to a greater extent.

28
THANX Any Questions?