1.5-side boundary labeling chun-cheng linnational chiao tung university sheung-hung poonnational...
Post on 19-Dec-2015
217 views
TRANSCRIPT
1.5-side Boundary Labeling
Chun-Cheng Lin National Chiao Tung UniversitySheung-Hung Poon National Tsing Hua UniversityShigeo Takahashi The University of TokyoHsiang-Yu Wu The University of TokyoHsu-Chun Yen National Taiwan University
Boundary labeling (Bekos et al., GD 2004)
(Bekos & Symvonis, GD 2005)
Type-opo leaders Type-po leders Type-s leaders
Min (total leader length or total bend number)s.t. #(leader crossing) = 0
1-side, 2-side, 4-side
sitelabel
leader
Variants
Polygons labeling (Bekos et. al, APVIS 2006)
Multi-stack boundary labeling (Bekos et. al, FSTTCS 2006)
1.5-side Boundary Labelingtype-opo: direct leader vs. indirect leader
Annotation system for wordprocessing S/W
#1#2#3#4
#5
#6
indirectleader
#1
#2#3#4
#5
#6
direct leader
Problem Setting
(labelSize, labelPort, Objective)
#1
#2
Fixed-position port (FP)Fixed-ratio port (FR) Sliding port
#1
#2
#1
#2
Problem Setting
(labelSize, labelPort, Objective)
Min (total bend num)(TBM for short)
Min (total leader length)(TLLM for short)
#1
#2
#3
#4
#4
#1
#2
#3#1
#2
#3
#3
#1#2
#(bends) = 6 #(bends) = 2longer length shorter length
Assumptions
All the parameters are integers
No two sites with the same x- or y- coordinate
Map height = label height sum
Legal leader
pj
pi
map label
Aleft
Aright
pi–1
j
i
pj+1
pj
pi
map label
Aleft
Arightpi–1
j
i
pj+1
#1
#2#3#4
#5
#6
Our Contributions
(LabelSize, LabelPort, Objective) time reference
(uniform, FR/sliding, TLLM) O(n log n) Thm 1
(uniform, FP, TLLM) O(n5) Thm 2
(uniform,FR/FP/sliding,
TBM) O(n5) Thm 2
(nonuniform,FR/FP/sliding,
TLLM) NP-complete* Thm 3
(nonuniform,FR/FP/sliding,
TBM) NP-complete* Thm 4
* Pseudo-polynomial time algorithms and fixed-parameter algorithms are designed for those intractable problems.
Solved all the problems of all the combinations of (LabelSize, LabelPort, Objective).
(LabelSize, LabelPort, Objective) time reference
(uniform, FR/sliding, TLLM) O(n log n) Thm 1
(uniform, FP, TLLM) O(n5) Thm 2
(uniform,FR/FP/sliding,
TBM) O(n5) Thm 2
(nonuniform,FR/FP/sliding,
TLLM) NP-complete* Thm 3
(nonuniform,FR/FP/sliding,
TBM) NP-complete* Thm 4
Lemma 1. All direct leaders are optimal for the above concerned case.
(LabelSize, LabelPort, Objective) time reference
(uniform, FR/sliding, TLLM) O(n log n) Thm 1
p
U
BpB
U
p
B
lhB
lvB
h
n
leader l
p
B
|U|
(LabelSize, LabelPort, Objective) time reference
(uniform, FR/sliding, TLLM) O(n log n) Thm 1
(uniform, FP, TLLM) O(n5) Thm 2
(uniform,FR/FP/sliding,
TBM) O(n5) Thm 2
(nonuniform,FR/FP/sliding,
TLLM) NP-complete* Thm 3
(nonuniform,FR/FP/sliding,
TBM) NP-complete* Thm 4
Theorem 2. The above case can be solved by dynamic programming in O(n5) time.
(LabelSize, LabelPort, Objective) time reference
(uniform, FP, TLLM) O(n5) Thm 2
pb
pa
(c+b-a)-th
c-th
map label
)}1 , ,1() , ,1() ,1,() ,( {min
)},1 , ,1()1 ,1 ,() ,1,() ,( {min
), ,(min{
},,,0{,
},,,0{,
0
jcbjaSicjaiaSciaaSjcp
icbiaSjciajaScjaaSjcp
icp
iaijabji
iaijabji
ab
iia
S(a, b, c) =
// all direct leaders
// downward indirect leader
// upward indirect leader
// the solution of the problem with pa, pa+1, …, pb connected to label positions c to c+(b-a)+1
# = (b+a)+1
Theorem 2. The above case can be solved by dynamic programming in O(n5) time.
(LabelSize, LabelPort, Objective) time reference
(uniform, FP, TLLM) O(n5) Thm 2
)}1 , ,1() , ,1() ,1,() ,( {min
)},1 , ,1()1 ,1 ,() ,1,() ,( {min
), ,(min{
},,,0{,
},,,0{,
0
jcbjaSicjaiaSciaaSjcp
icbiaSjciajaScjaaSjcp
icp
iaijabji
iaijabji
ab
iia
S(a, b, c) =
// all direct leaders
// downward indirect leader
// upward indirect leader
// the solution of the problem with pa, pa+1, …, pb connected to label positions c to c+(b-a)+1
pb
pa
(c+b-a)-th
c-th
map label
Theorem 2. The above case can be solved by dynamic programming in O(n5) time.
(LabelSize, LabelPort, Objective) time reference
(uniform, FP, TLLM) O(n5) Thm 2
pb
pa+i+1
pa+i-1
pa+j
pa+j-1
pa
pa+i
(c+b-a)-th
(c+i)-th
(c+j-1)-th
c-th
(c+j+1)-th
(c+i+1)-th
map label
S(a+i+1, b, c+i+1)
S(a+j, a+i-1, c+j+1)
S(a, a+j-1, c)
(c+j)-th
)}1 , ,1() , ,1() ,1,() ,( {min
)},1 , ,1()1 ,1 ,() ,1,() ,( {min
), ,(min{
},,,0{,
},,,0{,
0
jcbjaSicjaiaSciaaSjcp
icbiaSjciajaScjaaSjcp
icp
iaijabji
iaijabji
ab
iia
S(a, b, c) =
// all direct leaders
// downward indirect leader
// upward indirect leader
// the solution of the problem with pa, pa+1, …, pb connected to label positions c to c+(b-a)+1
Theorem 2. The above case can be solved by dynamic programming in O(n5) time.
(LabelSize, LabelPort, Objective) time reference
(uniform, FP, TLLM) O(n5) Thm 2
pb
pa+j+1
pa+j
pa+i+1
pa+i-1
pa
pa+i
(c+b-a)-th
(c+i)-th
(c+j-1)-th
c-th
(c+i-1)-th
(c+j+1)-th
map label
S(a+j+1, b, c+j+1)
S(a, a+i-1, c)
(c+j)-th
S(a+i+1, a+j, c+i)
pb
pa+i+1
pa+i-1
pa+j
pa+j-1
pa
pa+i
(c+b-a)-th
(c+i)-th
(c+j-1)-th
c-th
(c+j+1)-th
(c+i+1)-th
map label
S(a+i+1, b, c+i+1)
S(a+j, a+i-1, c+j+1)
S(a, a+j-1, c)
(c+j)-th
)}1 , ,1() , ,1() ,1,() ,( {min
)},1 , ,1()1 ,1 ,() ,1,() ,( {min
), ,(min{
},,,0{,
},,,0{,
0
jcbjaSicjaiaSciaaSjcp
icbiaSjciajaScjaaSjcp
icp
iaijabji
iaijabji
ab
iia
S(a, b, c) =
// all direct leaders
// downward indirect leader
// upward indirect leader
// the solution of the problem with pa, pa+1, …, pb connected to label positions c to c+(b-a)+1
(LabelSize, LabelPort, Objective) time reference
(uniform, FR/sliding, TLLM) O(n log n) Thm 1
(uniform, FP, TLLM) O(n5) Thm 2
(uniform,FR/FP/sliding,
TBM) O(n5) Thm 2
(nonuniform,FR/FP/sliding,
TLLM) NP-complete* Thm 3
(nonuniform,FR/FP/sliding,
TBM) NP-complete* Thm 4
Total Discrepancy Problem is NP-complete job Ji {J0, J1, …, J2n}
Execution time length li , where I0 < I1 < … < l2n
Preferred midtime M = (l0 + l1 + … + l2n) /2
For a planned scheduleActual midtime of Ji = mi()
Min ( |m0() – M| + |m1() – M| + … + |m2n() – M| + |m2n+1() – M’|)
Properties for the optimal schedule opt
No gaps between two jobs
m0(opt) = M
| {Ji : mi < M } | = | {Ji : mi > M } |
opt = An, An-1, …, A1, J0, B1, B2, …, Bn where {Ai, Bi} = {J2i-1, J2i}
0 M
J0J1 J2J3 J4
J0 J1 J2 J3 J4
Theorem 3. Total Discrepancy Problem L(nonuniform, FR/FP/sliding, TLLM).
(LabelSize, LabelPort, Objective) time reference
(nonuniform,FR/FP/sliding,
TLLM) NP-complete* Thm 3
0 M
J0J1 J2J3 J4
(LabelSize, LabelPort, Objective) time reference
(uniform, FR/sliding, TLLM) O(n log n) Thm 1
(uniform, FP, TLLM) O(n5) Thm 2
(uniform,FR/FP/sliding,
TBM) O(n5) Thm 2
(nonuniform,FR/FP/sliding,
TLLM) NP-complete* Thm 3
(nonuniform,FR/FP/sliding,
TBM) NP-complete* Thm 4
Theorem 4. Subset Sum Problem L(nonuniform, FR/FP/sliding, TBM).
(LabelSize, LabelPort, Objective) time reference
(nonuniform,FR/FP/sliding,
TBM) NP-complete* Thm 4
pn+1
pn+2
hmin
Subset Sum ProblemInput: A = {a1, …, an} and a num B = (a1 + … + an)/2
Question: find a subset A’ A such that sum(elements in A’) = B
< hmin< hmin
h/2
(LabelSize, LabelPort, Objective) time reference
(uniform, FR/sliding, TLLM) O(n log n) Thm 1
(uniform, FP, TLLM) O(n5) Thm 2
(uniform,FR/FP/sliding,
TBM) O(n5) Thm 2
(nonuniform,FR/FP/sliding,
TLLM) NP-complete* Thm 3
(nonuniform,FR/FP/sliding,
TBM) NP-complete* Thm 4
* Pseudo-polynomial time algorithms and fixed-parameter algorithms are designed for those intractable problems.
Theorem 5 (pseudo-polynomial algorithm).The above two cases can be solved in O(n4h) time, where h is the map height.
)} , ,1() , ,1() ,1,() ,( {min
)}, , ,1())( ,1 ,() ,1,() ,( {min
), ,(min{
0
1
0
1
0},,,0{,
0
1
0
1
0},,,0{,
1
00
j
lla
i
lla
j
llaiaijabji
i
llaia
j
lla
j
llaiaijabji
i
jja
ab
iia
htbjaShtjaiaStiaaShtp
htbiaShhtiajaStjaaShtp
htp
(LabelSize, LabelPort, Objective) time reference
(nonuniform,FR/FP/sliding,
TLLM) NP-complete* Thm 3
(nonuniform,FR/FP/sliding,
TBM) NP-complete* Thm 4
S(a, b, t ) =
)}1 , ,1() , ,1() ,1,() ,( {min
)},1 , ,1()1 ,1 ,() ,1,() ,( {min
), ,(min{
},,,0{,
},,,0{,
0
jcbjaSicjaiaSciaaSjcp
icbiaSjciajaScjaaSjcp
icp
iaijabji
iaijabji
ab
iia
S(a, b, c) =
// all direct leaders
// downward indirect leader
// upward indirect leader
// the solution of the problem with pa, pa+1, …, pb connected to label positions c to c+(b-a)+1
// the solution of the problem with pa, pa+1, …, pb connected to y-coordinate t
(uniform label case)
Theorem 6 (fixed-parameter algorithm).The above two cases using k different label heights can be solved in O(nk+4) time.
Theorem 5. The above two cases can be solved in O(n4h) time.
Lemma 2. num( positions of each label using k different label heights ) = O(nk).
pf.Induction on k
Assume num(…(k-1) …) = O(nk-1)
Consider each label, which is the i-th label from the bottom
(LabelSize, LabelPort, Objective) time reference
(nonuniform,FR/FP/sliding,
TLLM) NP-complete* Thm 3
(nonuniform,FR/FP/sliding,
TBM) NP-complete* Thm 4
h = nk h = nk
type-k
type-1, …, type-(k-1)(i –1) labels usingtype-1, type-2, …, type-k
O(nk-1) positions
at most O(n)
Conclusion
(LabelSize, LabelPort, Objective) time reference
(uniform, FR/sliding, TLLM) O(n log n) Thm 1
(uniform, FP, TLLM) O(n5) Thm 2
(uniform,FR/FP/sliding,
TBM) O(n5) Thm 2
(nonuniform,FR/FP/sliding,
TLLM) NP-complete* Thm 3
(nonuniform,FR/FP/sliding,
TBM) NP-complete* Thm 4
* Pseudo-polynomial algorithms and fixed-parameter algorithms are designed for those intractable problems.
Solved all the problems of all the combinations of (LabelSize, LabelPort, Objective).