Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE,
Год журнала:
2017,
Номер
10255, С. 1025502 - 1025502
Опубликована: Март 8, 2017
To
improve
the
over-segmentation
and
over-merge
phenomenon
of
single
image
segmentation
algorithm,a
novel
approach
combing
Graph-Based
algorithm
T-junctions
cues
is
proposed
in
this
paper.
First,
a
method
by
L0
gradient
minimization
applied
to
smoothing
target
eliminate
artifacts
caused
noise
texture
detail;
Then,
initial
result
using
graph-based
algorithm;
Finally,
final
results
via
region
fusion
strategy
t-junction
cues.
Experimental
on
variety
images
verify
new
approach's
efficiency
eliminating
noise,segmentation
accuracy
time
complexity
has
been
significantly
improved.
IEEE Transactions on Image Processing,
Год журнала:
2020,
Номер
30, С. 1542 - 1555
Опубликована: Дек. 15, 2020
Morphology
component
analysis
provides
an
effective
framework
for
structure-texture
image
decomposition,
which
characterizes
the
structure
and
texture
components
by
sparsifying
them
with
certain
transforms
respectively.
Due
to
complexity
randomness
of
texture,
it
is
challenging
design
components.
This
paper
aims
at
exploiting
recurrence
patterns,
one
important
property
develop
a
nonlocal
transform
sparsification.
Since
plain
patch
holds
both
cartoon
contours
regions,
constructed
based
on
such
sparsifies
well.
As
result,
could
be
wrongly
assigned
component,
yielding
ambiguity
in
decomposition.
To
address
this
issue,
we
introduce
discriminative
prior
recurrence,
that
spatial
arrangement
recurrent
patches
regions
exhibits
isotropic
differs
from
contours.
Based
prior,
only
Incorporating
into
morphology
analysis,
propose
approach
Extensive
experiments
have
demonstrated
superior
performance
our
over
existing
ones.
Visual Intelligence,
Год журнала:
2023,
Номер
1(1)
Опубликована: Окт. 13, 2023
Abstract
Occlusion
relationship
reasoning
aims
to
locate
where
an
object
occludes
others
and
estimate
the
depth
order
of
these
objects
in
three-dimensional
(3D)
space
from
a
two-dimensional
(2D)
image.
The
former
sub-task
demands
both
accurate
location
semantic
indication
objects,
while
latter
needs
among
objects.
Although
several
insightful
studies
have
been
proposed,
key
characteristic
occlusion
reasoning,
i.e.,
specialty
complementarity
between
boundary
detection
orientation
estimation,
is
rarely
discussed.
To
verify
this
claim,
paper,
we
integrate
properties
into
unified
end-to-end
trainable
network,
namely
feature
separation
interaction
network
(FSINet).
It
contains
shared
encoder-decoder
structure
learn
complementary
property
two
sub-tasks,
separated
paths
specialized
sub-tasks.
Concretely,
path
image-level
cue
extractor
capture
rich
information
boundary,
detail-perceived
extractor,
contextual
correlation
acquire
refined
features
In
addition,
dual-flow
cross
detector
has
customized
alleviate
false-positive
false-negative
boundaries.
For
estimation
path,
scene
context
learner
designed
around
boundary.
stripe
convolutions
are
built
judge
decoder
supplies
interaction,
which
plays
role
exploiting
paths.
Extensive
experimental
results
on
PIOD
BSDS
ownership
datasets
reveal
superior
performance
FSINet
over
state-of-the-art
alternatives.
Additionally,
abundant
ablation
offered
demonstrate
effectiveness
our
design.
SIAM Journal on Imaging Sciences,
Год журнала:
2020,
Номер
13(3), С. 1179 - 1210
Опубликована: Янв. 1, 2020
Related
DatabasesWeb
of
Science
You
must
be
logged
in
with
an
active
subscription
to
view
this.Article
DataHistorySubmitted:
25
September
2019Accepted:
15
April
2020Published
online:
16
July
2020Keywordscartoon-texture
decomposition,
patch
recurrence,
regularization
methodAMS
Subject
Headings68T05,
68U10,
65D18Publication
DataISSN
(online):
1936-4954Publisher:
Society
for
Industrial
and
Applied
MathematicsCODEN:
sjisbi
