siggraph 2010 structure-based ascii art xuemiao xu, linling zhang, tien-tsin wong the chinese...

Post on 17-Dec-2015

215 Views

Category:

Documents

2 Downloads

Preview:

Click to see full reader

TRANSCRIPT

SIGGRAPH 2010

Structure-based ASCII Art

Xuemiao Xu, Linling Zhang, Tien-Tsin Wong

The Chinese University of Hong Kong

Since the 1860s, text art emerged…

Since the 1860s, text art emerged…

From the 1970s, ASCII art has been widely used…

From the 1970s, ASCII art has been widely used…

Today, ASCII art remains popular…

ASCII Art Classification

• Structure-based• Tone-based– Halftone approaches

Regarded as dithering

– O’Grady and Rickard [2008]

Dithering essentially

• Structure-based– Manual

Tedious

ASCII Art Classification

Automatic generation of structure-based ASCII art

• Tone-based– Halftone approaches

Regarded as dithering

– O’Grady and Rickard [2008]

Dithering essentially

• Arbitrary image content

Main Challenge

• Arbitrary image content

• Extremely limited character shapes• Restrictive placement of characters

Main Challenge

__)

Matching Strategies

• Character matching – Misalignment tolerance– Transformation awareness

Matching Strategies

• Character matching – Misalignment tolerance– Transformation awareness

• Image deformation – Increase the chance of matching – Avoid over-deformation

__)

• Character matching – Misalignment tolerance– Transformation awareness

• Image deformation – increase the chance of matching – Avoid over-deformation

Alignment-insensitive shape similarity metric

Constrained deformation

Matching Strategies

Vectorized polylines

Framework

InputRasterized imageCurrent best matched characters

Matching error map

^)

Current best matched characters

Framework

Matching error map

Good matching

Poor matching

_

;r ;

Current best matched characters

Framework

Matching error map

(')(_)

(_)

Current best matched characters

Framework

Matching error mapDeformation cost map Combined cost map

Combined cost mapCurrent best matched characters

Framework

Deformed image Optimal ASCII art

• Deformation cost of the vectorized images

Objective Function

E = DAISS . Ddeform

• Shape dissimilarity between ASCII and deformed images

Main Contribution

• Shape MatchingAlignment-Insensitive Shape Similarity (AISS) Metric

• Constrained Deformation Deformation Metric

• Matching requirements • Misalignment tolerance• Transformation awareness

• Scope • Pattern recognition and image analysis, e.g. OCR

AISSOCR

O O 6 9

• Misalignment tolerance

Log-polar diagram (5x12)

Design of AISS

log-polar diagram

Log-polar histogram

• Transformation awareness

Design of AISS

h

New sampling layout

Query Shape Context Our metric

Translation and scale invariant

Metrics Comparison (1)

• Transformation-invariant metrics

Over-emphasize overlapping

Metrics Comparison (2)

SSIMQuery Our metric RMSEafter blurring

• Alignment-sensitive metrics

Main Contribution

• Shape MatchingAlignment-Insensitive Shape Similarity (AISS) Metric

• Constrained Deformation Deformation Metric

Constrained Deformation

• Local deformation constraint

• Accessibility constraint

AA

BB

r’r’

rr

Local Deformation Constraint

B’B’

A’A’

Accessibility Constraint

Optimization

Corresponding ASCII art Input Vectorized image

Resolution=30X20 Resolution=20X15

Comparison

Input O’Grady & Rickard Our method

Test set 3:Test set 2:

Input By Artist Our Method O’Grady & Rickard Input By Artist Our Method O’Grady & Rickard

Test set 1:

Clarity

Artists 7.18

Our method 7.09

O’Grady & Rickard 4.15

User Study

Similarity

Artists 6.86

Our method 7.36

O’Grady & Rickard 4.42

Input By Artist Our Method O’Grady &

Rickard

More Results

Other Results

Other Results

Conclusion

• Mimic ASCII artists’ work by an optimization process

• Propose a novel alignment-insensitive shape similarity metric

- also benefits pattern recognition

• Propose a new deformation metric to control over-deformation

• Do not consider the stylish variation of line thickness within a font

• Do not handle proportional placement of characters

• Affected by the quality of the vectorization

Limitation

A A

Q&A

top related