<s>
Image	B-Algorithm
segmentation	I-Algorithm
strives	O
to	O
partition	O
a	O
digital	O
image	O
into	O
regions	O
of	O
pixels	O
with	O
similar	O
properties	O
,	O
e.g.	O
</s>
<s>
To	O
speed	O
up	O
segmentation	B-Algorithm
of	O
large	O
images	O
,	O
the	O
work	O
could	O
be	O
divided	O
among	O
several	O
CPUs	B-Device
.	O
</s>
<s>
However	O
,	O
regions	O
that	O
straddle	O
a	O
tile	O
border	O
might	O
be	O
split	O
up	O
or	O
lost	O
if	O
the	O
fragments	O
do	O
not	O
meet	O
the	O
segmentation	B-Algorithm
algorithm	O
's	O
minimum	O
size	O
requirements	O
.	O
</s>
<s>
An	O
alternative	O
exists	O
in	O
the	O
form	O
of	O
graph-based	O
segmentation	B-Algorithm
methods	O
.	O
</s>
<s>
The	O
possibility	O
of	O
stitching	B-Algorithm
together	O
independent	O
sub-images	O
motivates	O
adding	O
connectivity	O
information	O
to	O
the	O
pixels	O
.	O
</s>
<s>
In	O
2004	O
,	O
Felzenszwalb	O
introduced	O
a	O
segmentation	B-Algorithm
method	O
based	O
on	O
Kruskal	B-Algorithm
's	I-Algorithm
MST	I-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
Detecting	O
cycles	O
is	O
possible	O
in	O
near-constant	O
time	O
with	O
the	O
aid	O
of	O
a	O
disjoint-set	B-Algorithm
data	I-Algorithm
structure	I-Algorithm
.	O
</s>
<s>
The	O
complexity	O
of	O
the	O
algorithm	O
is	O
quasi-linear	O
because	O
sorting	O
edges	O
is	O
possible	O
in	O
linear	O
time	O
via	O
counting	B-Algorithm
sort	I-Algorithm
.	O
</s>
<s>
Instead	O
of	O
a	O
fixed	O
weight	O
threshold	O
,	O
an	O
initial	O
connected-component	B-General_Concept
labeling	I-General_Concept
is	O
used	O
to	O
estimate	O
a	O
lower	O
bound	O
on	O
the	O
threshold	O
,	O
which	O
can	O
reduce	O
both	O
over	O
-	O
and	O
undersegmentation	O
.	O
</s>
<s>
In	O
2017	O
,	O
Saglam	O
and	O
Baykan	O
used	O
Prim	O
's	O
sequential	O
representation	O
of	O
minimum	O
spanning	O
tree	O
and	O
proposed	O
a	O
new	O
cutting	O
criterion	O
for	O
image	B-Algorithm
segmentation	I-Algorithm
.	O
</s>
