Page 174 - 2024-Vol20-Issue2

170 |
Fig. 6. Optical encryption process.
Fig. 7. Schematic diagram of the DCPE-based encryption
scheme.
Here, (x, y) and (u, v) represent the image plane and Fourier
plane coordinates, respectively. The second step is to bound
the obtained Fourier spectrum with a statistically indepen-
dent CPM2, exp(2p jCPM2), and retrieve the resulting spectrum
once more FT
E(x, y) = FT [E1(u, v)CPM2 R(u, v)]
= [E1(u, v)exp(2p jCCPM2 R(u, v)) (6)
exp(-2p j(ux + uv))]dudv
where E(x, y) is the field of the final encryption image.
B. Proposed Decryption Scheme
Decryption is the inverse process of encryption and its aim is
to recover the input images from the encrypted image. Fig. 8
shows a block diagram of the proposed hybrid digital / optical
decryption (HDOD) scheme. The encryption and decryption
keys are identical, and the chaotic sequence generated during
decryption is consistent with that generated during encryption. cryption / decryption algorithms described in Section III. The
C. Denoising Convolution Neural Network
A quick and efficient denoiser convolution neural network
(DnCNN) based on the DL principle is used in this work to
Muttashar & Fyath
Fig. 8. Block diagram for hybrid optical / digital decryption
scheme.
Fig. 9. Architecture of the DnCNN.
address the issue that encrypted images are susceptible to
some assaults in real applications. The resolution of the re-
constructed images is increased by applying CNN, which also
increases the algorithm’s security robustness. The structure
of DnCNN consists of an input image layer with a patch size
(50) and four ”convolutional layers + ReLU”; for each one of
the first three layers, 64 filters (kernels) of size 3x3 are used
to generate 64 feature maps. Each layer produces a feature
map except the fourth layer, which does not have a ReLU but
a stride = one and a ”Regression layer.” The zero padding is
implemented to ensure that the feature map and input image
have the same size. In order to distinguish the difference be-
tween a noisy and a clean image, the DnCNN is used in our
system. Fig. 9 illustrates the architecture of the used DnCNN.
IV. SIMULATION RESULTS
This section presents simulation results for the proposed en-
performance is evaluated using conventional encryption qual-