Digital
watermarking
is
a
widely
used
technique
for
embedding
information
into
digital
media
to
protect
intellectual
property
rights.
However,
watermarks
are
vulnerable
various
types
of
malicious
attacks.
In
this
paper,
we
propose
an
AES-based
scheme
further
improve
the
security
and
robustness
existing
watermark
technique.
The
proposed
involves
original
host
image
using
technique,
then
encrypting
embedded
AES
algorithm.
watermarked
subjected
series
attacks
decrypted
same
To
assess
performance
approach
in
terms
robustness,
it
implemented
tested
on
set
images.
experimental
findings
demonstrate
that,
while
maintaining
low
distortion
rate,
offers
high
against
can
be
considered
as
effective
secure
solution
protecting
applications.
Image
captioning
involves
generating
a
natural
language
description
that
accurately
represents
the
content
and
context
of
an
image.
To
achieve
this,
image
utilises
various
machine
learning
techniques
fields,
such
as
computer
vision
processing.
In
field
captioning,
lot
advances
have
been
made
with
encoder-decoder
models
reinforcement
algorithms.
However,
there
are
still
problems
imbalance
between
testing
training,
only
handles
single
comparator
metrics
CIDEr,
SPICE,
BLEU
could
not
perform
better
in
multiple
at
once.
Which
is
why
lack
diversity
can
be
seen
generated
captions.
This
idea
proposes
general
technique
for
collaborative
updating
bridge
gap
evaluation
measures
test
to
produce
captions
more
human-like.
increase
precision
captions,
approach
using
compiled
reward
system
considers
compare
sentence
provided
sentences.
We
will
evaluate
model's
performance
process
on
standard
datasets
like
MS
COCO.
A
robust
watermarking
scheme
is
proposed
in
this
paper.
The
uses
discrete
wavelet
transformations
conjunction
with
Hessenberg
decomposition
and
singular
value
to
decompose
the
cover
image
watermark
image.
Every
single
of
instantly
integrated,
utilising
optimal
scaling
factor,
into
solitary
component
after
application
decomposition.
To
make
attacked
attack-free
high-quality,
a
reconstruction
technique
integrated
scheme,
considering
various
processing
geometrical
attack
scenarios.
more
resistant
noise-based
attacks,
according
experimental
findings
computed
terms
several
performance
metrics,
including
peak
signal-to-noise
ratio
(PSNR),
structure
similarity
index
(SSIM),
bit
error
rate
(BER),
normalised
absolute
(NAE).
International Journal on Semantic Web and Information Systems,
Journal Year:
2024,
Volume and Issue:
21(1), P. 1 - 24
Published: Dec. 10, 2024
Face
super-resolution
generates
high-resolution
face
images
from
low-resolution
inputs,
supporting
recognition
in
challenging
environments.
While
deep
learning
methods,
like
Stable
Diffusion
and
origin
Transformer-based
framework,
have
advanced
super-resolution,
they
require
heavy
computation,
making
them
difficult
for
tasks
recognition.
However,
accuracy
only
needs
of
size
112x112,
reducing
the
necessity
extremely
large
outputs.
Existing
methods
often
rely
on
convolutional
attention,
limiting
receptive
fields
performance.
To
address
this,
we
propose
SETFSR,
a
Semantically
Guided
Efficient
Attention
Transformer
Super-Resolution.
Our
model
leverages
efficient
self-attention
global
feature
extraction
incorporates
parsing
constraints
structural
accuracy.
Experiments
Helen,
CelebA,
FFHQ
datasets
show
that
SETFSR
outperforms
state-of-the-art
models
PSNR,
SSIM,
identity
preservation.
