Traitement du signal,
Год журнала:
2023,
Номер
40(1), С. 227 - 234
Опубликована: Фев. 28, 2023
Infrared
thermal
image
makes
the
target
have
certain
degree
of
recognition
by
reflecting
radiation
information
emitted
target,
which
effectively
compensates
loss
visible
light
in
harsh
imaging
environment.Contour
extraction
effect
area
using
traditional
Canny
algorithm
is
not
good,
because
contour
gradient
change
infrared
obvious.At
same
time,
threshold
most
algorithms
needs
to
be
set
manually,
greatly
affected
subjective
factors,
and
processing
efficiency
low.Therefore,
this
paper
studied
combining
detection
with
matching
correction.First,
introduced
feature
based
on
grid
motion
statistics,
converted
smoothness
constraint
into
thus
replacing
number
extended
points
acquisition
features
better
performance
filtering
false
other
neighborhood
statistical
points.Second,
results
obtained
previous
section,
proposed
a
method
descriptors,
combined
extracted
semantic
attributes
each
image,
distinguishing
subtle
differences
between
sub-categories.Finally,
experimental
verified
effectiveness
method.
IEEE Transactions on Computational Imaging,
Год журнала:
2024,
Номер
10, С. 385 - 398
Опубликована: Янв. 1, 2024
Existing
image
fusion
methods
focus
on
aggregate
features
from
different
modalities
into
a
clear
and
comprehensive
image.
However,
these
solutions
ignore
the
importance
of
gradient
features,
which
results
in
smooth
performance
contrast
information
fused
images.
In
this
paper,
an
edge
enhancement
method
for
infrared
visible
is
proposed,
named
EgeFusion.
First,
source
images
are
decomposed
series
base
detail
layers
through
simple
weighted
least
squares
filter.
Next,
sub-window
variance
filter
proposed
layers.
For
layer,
strategy
that
combines
visual
saliency
mapping
with
idea
adaptive
weight
assignment
designed.
The
effectively
assigns
globally,
thus
providing
more
valuable
about
region
interest
Finally,
reconstructed
reverse
to
obtain
results.
experimental
show
has
significantly
enhanced
images,
makes
it
easier
human
eye
system
interest.
Compared
other
state-of-the-art
methods,
EgeFusion
superior
quality
acceptable
visible,
multi-focus,
as
well
multi-modal
medical
fusion.
More
importantly,
our
approach
achieves
improvements
object
detection.
IEEE Sensors Journal,
Год журнала:
2024,
Номер
24(4), С. 4920 - 4934
Опубликована: Янв. 3, 2024
Infrared
and
visible
image
fusion
can
effectively
integrate
the
advantages
of
two
source
images,
preserving
significant
target
information
rich
texture
details.
However,
most
existing
methods
are
only
designed
for
well-illuminated
scenes
tend
to
lose
details
when
encountering
low-light
because
poor
brightness
images.
Some
incorporate
a
light
adjustment
module,
but
they
typically
focus
on
enhancing
intensity
neglect
enhancement
color
feature,
resulting
in
unsatisfactory
visual
effects
fused
To
address
this
issue,
paper
proposes
novel
method
called
EV-fusion,
which
explores
potential
detail
features
images
improve
perception
Specifically,
an
unsupervised
module
is
that
restores
texture,
structure
by
several
non-reference
loss
functions.
Then,
devised
enhanced
infrared
image.
Moreover,
salient
object
feature
we
propose
bilateral-guided
salience
map
embedding
into
Extensive
experiments
demonstrate
our
outperforms
state-of-the-art
methods.
