STEGANOGRAPHIC METHOD FOR COLOR IMAGES USING VARIANCE FIELD ESTIMATION AND SCALING FACTOR FOR ENERGY CONSERVATION - Theoretical and Computational Research in the 21st Century - page 95

sub-band LH of the cover image using a redundancy of the approaches
algorithm. This algorithm is used because it is easy to implement, reduces
the computational cost, and provides good fi ltering results [7, 23-25]. The
sub-band LH is chosen because it provides the information of the edges
and the details of the cover image which are good locations to hide the
data and besides with the preprocessing step there are more variations in
the pixel intensities, increasing the embedding capacity and improving the
steganography security.
Step 2a: Apply second-level Daubechies db4 DWT This step is used
to hide data. Let us use the subband LH obtained from the wavelet coef-
fi cients of the subband LL from the second decomposition of the resulting
image of the step 2. The decomposition level and the Daubechies db4 are
selected according with the extensive experimental results realized with
different wavelet families under objective criteria [8, 26-29].
Stage B: This stage allows having more hiding capacity in the pro-
posed method because it provides more decomposition levels to hide in-
formation [8].
Step 2a: Apply second-level Daubechies db4 DWT This step is the
same as discussed earlier.
The next steps are applied in both outputs of the stages A and B provid-
ing more hiding capacity [8].
Step 3: Standard deviation computation The embedding capacity for
each pixel is determined by the local complexity of the cover color image
providing a good visual quality as well as embedding a large amount of
secret messages [6]. From the wavelet decomposition, let us use the stan-

n

∑ dard deviation

σ = ( xxn − )

2

/

of the subband LH obtained from the

k m

m = 1

wavelet coeffi cients from the subband LL of the second decomposition of
the cover color image using a 3 × 3 kernel, where

x is an element of the

n

∑ =

current kernel k ,

x = x / n

is the mean value of current kernel, and

m

m 1

n is the number of elements in the sample [8, 27, 30].
Step 4: Threshold computation The pixels are classifi ed using a
threshold based on the local complexity of the standard deviation in the
cover image . The threshold

= 9

( ) ∑

∑

− s − s

T = α ⋅ 2 2

is used to select the pix-
els whose values are considered as places to hide data [31, 32], where T
denotes a threshold value for discriminating signal-dominant scales from

s

s s