Page 18 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

6 Chapter 1 Congruence of deep learning in biomedical engineering

Figure 1.5 Fire module with the hyperparameters: s1 1 ¼ 3, e1 1 ¼ 4, and e3 3 ¼ 4[37]. From Forrest N.
Iandola, Matthew W. Moskewicz, Khalid Ashraf, Song Han, William J. Dally and Kurt Keutzer, SqueezeNet: AlexNet-level
accuracy with 50x fewer parameters and <0.5MB model size, conference paper, ICLR 2017,arXiv:1602.07360 [cs.CV].
Steganography in a picture is confined in a spatial territory
and change zone. In the spatial region system the conventional
picture space authentically changes position. It is where informa-
tion storage is performed endlessly along with direct estimation
of the pixel of the spread picture. The effect of the message is
not discernible on the spread picture. In the change zone, meth-
odologies rely upon modifying the Fourier difference in an image;
the hidden development changes the spread picture into another
space. The changed coefficients are used to cover secret
messages. These changed coefficients are changed into spatial
space to obtain the stego picture. Essentially, it is a lossless
method and the additional substance commotion sum can be
taken in the photos quickly.
There are various types of parameters (iteration size, batch
size, optimization algorithm, learning rate, momentum, weight
decay, regularization, and dropout) used to measure the perfor-
mance of the procedure, which are described in Table 1.2.
Transform domain techniques are more advantageous than
spatial domain strategies because they are utilized for concealing
the data in the region of the picture that has a reduced amount
of cover in compacting, editing, and picture preparing. Transform
domain strategy does not show up in the picture and surpasses
lossless and lossy interpretations. Most of the stenographic frame-
works perceived currently basically take a shot at some technique
for change space. These methods are utilized to shroud data that
are in significant pieces of the spread picture. It makes them more
solid to events. One procedure is to utilize the Fourier and cosine
changes, for example, discrete Fourier transform (DFT) or discrete