Image-Guided Maze Construction 논논 논논논 논논논논논 논논논논 논논논 논논논 2007.10.18 1
Image-Guided Maze Construction
논문 세미나고려대학교 그래픽스 연구실
윤종철2007.10.18
1
목차 Abstract Introduction Maze basics Related work Maze textures
◦ Directional mazes
◦ Spiral and vortex mazes
◦ Random mazes
◦ User-defined lines
User-specified solution paths Additional effects
◦ Tone reproduction
◦ Foreshortening
Implementation and results Conclusions and Future Work
2
Abstracta set of graphical and combinato-
rial algorithms for designing mazes based on images
3
Introduction
4
IntroductionMazes and labyrinths have en-
joyed a long, venerable tradition in the history of art and design.
They have been used as pure visual art, as architectural deco-ration, and as cultural and reli-gious artifacts
An interactive application that lets a designer author a maze at a high level.
5
Related workVortex maze construc-
tion [Jie Xu 2006]◦ Technique for drawing abstract
geometric mazes based on ar-rangements of vortices
Organic Labyrinths and Mazes [Pedersen 2006]◦ Single paths with no branch
6
Maze basicsKruskal’s algorithm
◦ 1. graph 의 모든 edge 를 가중치로 오름차순 정렬
◦ 2. 가중치가 가장 작은 곳에 edge 를 삽입 , 이때 cycle 을 형성하는 edge 는 삽입할 수 없으므로 다음 가중치가 작은 edge 삽입
◦ 3. n-1 개의 edge 를 삽입할 때까지 2 반복◦ 4. edge 가 n-1 개가 되면 spanning tree 완성
7
Maze basics
Kruskal’s algorithm◦Cycle 판별
a 와 b 라는 노드가 선택되었을 때 , 1) a 와 b 가 서로 다른 집합이면 a 와 b 는 연결해도
cycle 이 생기지 않는다 . 2) a 와 b 가 서로 같은 집합에 속해 있다면 a 와 b 를
연결하면 cycle 이 생긴다 . 1 번의 경우 edge 를 연결하고 a 가 속한 집합과 b 가
속한 집합을 합쳐주고 , 2 번의 경우에는 edge 를 선택하지 않는다 .
8
Maze basics
9
Maze basicsex) To bias maze construction
◦0<a<b<1◦Assign horizontal walls weights cho-
sen from the interval [0,b], and ver-tical walls weights from [a,1]
Horizontal walls are therefore more likely to be deleted first
10
11
12
13
Perfect maze : When each of these paths is unique then the maze contains no cycles and is called perfect
14
Segmentation
15
not automate the segmenta-tion,Intelligent Scissors [Mortensen 1995]
Maze texturesMaze textures
◦Directional mazes◦Spiral and vortex mazes◦Random mazes◦User-defined lines
16
Maze textures
(a) directional region(b) spiral region,(c) random region(d) user-defined lines
17
18
Vortex texture
19
20
21
Random texture
22
Random texture
23
User-specified solution paths
24
User-specified solution paths
25
User-specified solution paths
26
User-specified solution paths
27
User-specified solution paths
28
A B C A
B
C 1
11
1
11
2
2
User-specified solution paths
29
α
β
A B C A
B
C 2
2
1 1
1
1
>(O)
User-specified solution paths
30
User-specified solution paths
31
User-specified solution paths
32
Avoidance direct passages
33
Additional effectsTone reproductionForeshortening
34
Tone reproduction
35
Tone reproductionLightness G = (S-W)/S
◦ S : the spacing between the centres of the lines◦ W : line Width
◦ P : passage width S-W
36
S
W
P
Tone reproductionWe define
◦minimum line width Wmin
◦minimum passage width Pmin
◦The largest acceptable line spacing Smax
The darkest tone : ◦S = Smax, S−W = Pmin
◦lightness Gmin = Pmin/Smax Similarly, the lightest available
tone is Gmax = (Smax−Wmin)/Smax
37
Tone reproductionBoth passage width and line
width are minimized◦Gthresh = Pmin / Pmin+Wmin
◦G’ is computed by mapping G into the range [Gmin,Gmax]
When G’<=Gthresh, S=Pmin/G’, W=Pmin(1-G’)/G’
When G>Gthresh, S=Wmin(1-G’), W=Wmin
38
Foreshortening
39
40
Implementation and re-sultsC++, CGAL libraryDesign process requires only a
few minutes of user interactionMulti-thread
41
Results
42
Results
43
Results
44
Results
45
Conclusions and Future WorkA system for designing mazes
that are stylized line drawings of images
The perfect mazes we construct here are but one possible maze topology. ◦It is also possible to construct mazes
containing cycles, or indeed mazes with no dead ends at all
Mathematical structure and hu-man psychology 46
END47