Page 23 - 2023-Vol19-Issue2

19 | Taheri

quency parts are upper left corner of DCT coefficient matrix the deficient, and contaminated and contortion of the data in-
and frequency range increments slantingly from upper passed
on corner to base right corner of DCT coefficient matrix [7]. formation can be made once more into a total model through
Integer wavelet transform strategy and hereditary calculation
based image steganography have improved results for robust- their own great adaptation to non-critical failure. Assume
ness against visual attacks contrasted with discrete wavelet
transform [8],[9]. the watermark signal W is binary images of N×N, as pixel

The image transformed by wavelet, a large portion of its values. The training process of neural networks is as follows:
energy is moved in low-frequency sub-band, in the event that
watermark is embedded straightforwardly in low frequency W = {W (i, j), 1 ? i, j ? N, w(i, j) ? {0, 1}}
sub-band, and the straightforwardness of image containing
watermark will ignore. In the event that watermark is embed- The watermark signal is coded. First 0 of binary image
ded straightforwardly in high frequency sub-band, a ton of
high frequency data is lost while containing watermark image watermark is changed to -1, because the values 0 is not ap-
in the wake of sifting, and the calculation robustness will de-
cline. The correlation between’s coefficients is bigger in low plicable in hopfield neural network, 1 is not changed, then
frequency sub-band, and then it separated further by discrete
cosine transform (DCT). After DCT, most of the energy in W’ is gotten; Then W’ will be divided into each overlapping
low frequency sub-band focused on few low frequency coeffi-
cients, so a large portion of the energy is concentrated about size for the n1× n2 sub-blocks, among which the m piece
the entire image. The image changed greater assuming these is,Wm' (1 '
coefficients corrected inconsistent; in this manner, it ought ? m ? N × N ), will be last translated into one-
to ensure that these coefficients didn’t alter. Since the high n1 n2 Wm
coefficients isn’t delicate in the eye, the high frequency parts
in low frequency sub-band are the ideal provincial of embed- dimension signal R, based on certain key.The hopfield net-
ded watermark [7], the inconsistencies about watermarking
robustness and straightforwardness can be tackled [10]. work training mode set R is gotten. The structure of hopfield

The proposed algorithm presents hybrid integer wavelet network is shown in Fig.1.
transform and discrete cosine transform based watermarking
procedure to acquire expanded indistinctness and robustness R = rm(k), 1 ? k ? n1 × n2|rm((i - 1) × n1 + j) = w”(i, j)
contrasted with IWT-DCT based watermarking strategy. The (1)
IWT is a type of DWT (Discrete Wavelet Transform) that
coefficients are all quantized to closest integer values. The hopfield network’s neuron number as n1 × n2 is built
and the weights of network node are initialized. The network
weights are adjusted according to certain rules, and the mode
is stored in network. The network weights adjusted well as
KEY of scrambling, and KEY is saved in order to extract

''

watermark. The output of network is vector R or matrix W ,
this is the final scrambled watermark stream.

II. SCHEME DESIGN Fig. 1. Hopfield Network Structure

A. Neural Network Training
There are numerous types of strategies for the watermark

signal preprocessing. The watermark is upset to cause the
watermark calculation to have great robustness, so the water-
mark will have arbitrariness. The examination shows that the
capability of watermarking calculation’s shearing [11] and
controlling will be further developed through significant wa-
termark image is mixed. There are just adding and deduct
activity in relative transform. So, the estimation for taking
modulus is stayed away from. The activity is quick [12].
There are succinct scientific articulations in its inverter change,
and process duration of cycle isn’t handled in the decoding
system. The image is totally mixed exclusively for 10 time’s
iterative cycle. So, the relative transform is utilized in the
watermark signal pretreatment. The Hopfield network as a
neural network possesses the cooperative memory capability,