Legible Compact Calligrams Changqing Zou 1 2 1
Legible Compact Calligrams Changqing Zou 1, 2 1 Simon Fraser University Junjie Cao 3 Warunika Ranaweera 1 Ibraheem Alhashim 1 Ping Tan 1 Alla Sheffer 4 Hao Zhang 1 2 Hengyang Normal University 3 Dalian University of Technology 4 University of British Columbia
Legible Compact Calligram flamingo 2
Legible Compact Calligram Legible: readable 3
Legible Compact Calligram sedan 4
Artist creations Calligrams created by professional artists 5
Fully automatic calligram generation? automatically umbrella 6
Requirement #1: Letters convey the shape Boundary conformation + filling the interior 7
Requirement #2: Legibility elephant Legibility of individual letters AND the word(s) 9
Challenges Challenge 1: Automatic co-placement of a set of letters 9
Challenges Challenge 2: Co-deformation of a set of letters 10
Challenges 0. 60 0. 95 Legibility measure for deformed letters Calligram 1 > < ? Calligram 2 = Legibility measure for deformed word(s) Challenge 3: Legibility measure is a new problem 11
Key Insight Protrusion feature for letter CO-placement 12
Key technique A reliable legibility measure 13
Outline p Pipeline p Methodology detail p Results p Limitation 14
Our solution Alignment Deformation Coarse-to-fine strategy based pipeline 15
Outline p Methodology detail • Letter alignment • Letter deformation • Legibility learning p Results p Limitation p Conclusion 16
Initial letter alignment kangaroo 17
Initial letter alignment kangaroo 18
Initial letter alignment kangaroo p Type and orientation compatibility p Location compatibility p Non- inverse letter sequence 19
Refined alignment 20
Objective function for refining alignment Letter legibility Shape conformation Word legibility üLetter legibility: how legible a letter with aspect ratio changed is üShape conformation: how closely the outer hull and interior coverage of the letters fit the input shape ü Word legibility: how similar the statues of all the letters are 21
Outline p Pipeline p Methodology detail • Letter alignment • Letter deformation • Legibility learning p Results p Limitation 22
Legibility-driven deformation Letter deformation based on the movements of control points 23
Outline p. Methodology detail • Letter alignment • Letter deformation • Legibility learning [Liang et. al 2014] p Results p Limitation p. Conclusion 24
Steps of legibility score learning Training Data: 30, 000+ deformed shapes for 26 lowercase Latin letters 80% 1500+ samples Thickening user-drawn skeletons 20% Wrapping 25
Legibility feature p Step 1: Separate each letter into its strokes p Step 2: Represent each stroke by its skeleton and thickness profile p Step 3: Concatenate the features from all strokes P θ Directions of skeleton points P Locations of skeleton points Thickness of strokes 26
Legibility measure learning Most open Least open Most smiling Least open Most formal Least formal Relative attribute ranking, [LIANG et al. 2014] 27
Outline p Pipeline p Methodology detail p Results p Limitation p Conclusion 28
Results: animal skunk 29
Results: animal crocodile 30
Results: animal whale 31
Results: wild images highheel 32
Results: wild images dreamliner 33
Results: wild images disneyland 34
Input variations: same outline image 35
Input variations: same word 36
Parameter variations Protrusion correspondence Orientation consistence Results from different parameter settings 37
User study: overall quality Result from our algorithm Result from artist ü 2 Equality ü 9 pairs of algorithmic and artist-designed calligrams ü 30 participants; 270 responses ü 50. 74% favored our method over artist’s designs or indicated that they are of equality (9. 56%) 38
Word animals vs. Professional artist Fully automatic 39
Word animals vs. Professional artist Fully automatic 40
User study: algorithmic vs. manual Result from our algorithm Result from users ü Equal legible ü 9 pairs of calligrams ü 30 participants; 270 responses ü 41. 44% responses ranked our method ahead of all human creators 41
User study: algorithmic vs. manual User 1 User 2 Automatic 42
Outline p Pipeline p Methodology detail p Results p Limitation p Conclusion 43
Limitations 44
Limitations Select/obtain a well-matched input automatically? 45
Conclusions p The first fully automatic method for legible, compact calligram generation. p Legibility measure can be potentially useful for other text layout problems. 46
Thanks ! Q&A Acknowledge: Mr. Dan Fleming, SIGGGRAPH reviewers, user study participates, NSERC Canada.
- Slides: 47