26
Generation of Chinese Character Based Generation of Chinese Character Based on Human Vision and Prior Knowledge o on Human Vision and Prior Knowledge o f Calligraphy f Calligraphy 报报报 报报 报报报 报报 报报 报报 报报报报 报报报报 报报报 、、、 报报 报报 报报报报 报报报报 报报报 、、、 报报 报报报报报报报报报报报报报报 报报 报报报报报报报报报报报报报报 NLP & CC 2012: 报报报报报报报 报报报报报报报报报报报报 报报报报报报报 报报报报报报报报报报报报

Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

  • Upload
    murray

  • View
    45

  • Download
    3

Embed Size (px)

DESCRIPTION

NLP & CC 2012:. Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy. 基于人类视觉和书法先验知识的汉字自动生成. 报告人: 史操 作者: 史操、肖建国、贾文华、许灿辉 单位: 北京大学计算机科学技术研究所. Abstract. - PowerPoint PPT Presentation

Citation preview

Page 1: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Generation of Chinese Character Based on HumaGeneration of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphyn Vision and Prior Knowledge of Calligraphy

报告人: 史操报告人: 史操作者: 史操、肖建国、贾文华、许灿辉作者: 史操、肖建国、贾文华、许灿辉单位: 北京大学计算机科学技术研究所单位: 北京大学计算机科学技术研究所

NLP & CC 2012:

基于人类视觉和书法先验知识的汉字自动生成基于人类视觉和书法先验知识的汉字自动生成

Page 2: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Abstract本文根据中国书法先验知识对汉字笔画及结构进本文根据中国书法先验知识对汉字笔画及结构进

行建模,提出一种行建模,提出一种五层五层框架用于表征框架用于表征笔画笔画及及部件部件间间的的相关性相关性。通过分析书法家特有的书法技巧,拆分书。通过分析书法家特有的书法技巧,拆分书法字为本文提出的五层框架提供充足的笔画及部件法字为本文提出的五层框架提供充足的笔画及部件样样本本 。根据 。根据 MarrMarr 提出的对人类视觉的若干假设,本提出的对人类视觉的若干假设,本文提出了文提出了可量化可量化的视觉美学准则用于指导汉字自动生的视觉美学准则用于指导汉字自动生成。使用贝叶斯统计原理,可将整个汉字生成过程描成。使用贝叶斯统计原理,可将整个汉字生成过程描述为一个述为一个贝叶斯动态模型贝叶斯动态模型。模型中。模型中状态方程状态方程控制控制对笔画、部件的线性变换,并且将本文提出的视觉美对笔画、部件的线性变换,并且将本文提出的视觉美学准则应用于学准则应用于观测方程观测方程之中。之中。

Page 3: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Previous Work

1. Handwriting Imitation1. Handwriting Imitation

2. Samples Based Character Generation2. Samples Based Character Generation ( Font Companies )( Font Companies )

Page 4: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Previous Work

1. Handwriting Imitation1. Handwriting Imitation

Page 5: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Previous Work

2. Samples Based Character Generation2. Samples Based Character Generation

Yan Zhenqing’s CalligraphyYan Zhenqing’s Calligraphy Generated by our algorithmGenerated by our algorithm

Page 6: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Previous Work

2. Samples Based Character Generation2. Samples Based Character Generation

Yan Zhenqing’s CalligraphyYan Zhenqing’s Calligraphy Generated by our algorithmGenerated by our algorithm

Page 7: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Previous Work

2. Samples Based Character Generation2. Samples Based Character Generation

Yan Zhenqing’s CalligraphyYan Zhenqing’s Calligraphy Generated by our algorithmGenerated by our algorithm

Page 8: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Modeling Prior Knowledge of Calligraphy

Character

Radical

Different Formof Radical

Stroke

Different Formof Stroke

Five layers framework to represent Chinese characterFive layers framework to represent Chinese character

Page 9: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Modeling Prior Knowledge of Calligraphy

Modeling Stroke and DFSModeling Stroke and DFS

Page 10: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Modeling Prior Knowledge of Calligraphy

Page 11: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Modeling Prior Knowledge of Calligraphy

Modeling Radical and DFRModeling Radical and DFR

(a) (b) (c) (d)(a) (b) (c) (d)

Page 12: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

David Marr's “Vision”David Marr's “Vision”investigates the investigates the spatial arrangementspatial arrangement of objects in an image of objects in an image1. Average local intensity;1. Average local intensity;2. Average size of similar objects;2. Average size of similar objects;3. Local density of the objects;3. Local density of the objects;4. Local orientation of the objects;4. Local orientation of the objects;5. Local distances associated with the spatial arrangement of 5. Local distances associated with the spatial arrangement of

similar objects;similar objects;6. Local orientation associated with the spatial arrangement of 6. Local orientation associated with the spatial arrangement of

similar objects.similar objects.

Page 13: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

David Marr's “Vision”David Marr's “Vision”investigates the spatial arrangement of objects in an imageinvestigates the spatial arrangement of objects in an image1. Average local intensity;1. Average local intensity;2. Average size of similar objects;2. Average size of similar objects;3. Local density of the objects;3. Local density of the objects;

Different Formof Radical

Stroke

Different Formof Stroke

Different Formof Radical

Stroke

Different Formof Stroke

Different Formof Radical

Stroke

Different Formof Stroke

Different Formof Radical

Stroke

Different Formof Stroke

Different Formof Radical

Stroke

Different Formof Stroke

Page 14: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

David Marr's “Vision”David Marr's “Vision”investigates the spatial arrangement of objects in an imageinvestigates the spatial arrangement of objects in an image1. Average local intensity;1. Average local intensity;2. Average size of similar objects;2. Average size of similar objects;3. Local density of the objects;3. Local density of the objects;

Different Formof Radical

Stroke

Different Formof Stroke

Page 15: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

David Marr's “Vision”David Marr's “Vision”investigates the spatial arrangement of objects in an imageinvestigates the spatial arrangement of objects in an image1. Average local intensity;1. Average local intensity;2. Average size of similar objects;2. Average size of similar objects;3. Local density of the objects;3. Local density of the objects;

Different Formof Radical

Stroke

Different Formof Stroke

Page 16: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

David Marr's “Vision”David Marr's “Vision”investigates the spatial arrangement of objects in an imageinvestigates the spatial arrangement of objects in an image1. Average local intensity;1. Average local intensity;2. Average size of similar objects;2. Average size of similar objects;3. Local density of the objects;3. Local density of the objects;

Different Formof Radical

Stroke

Different Formof Stroke

Page 17: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

David Marr's “Vision”David Marr's “Vision”investigates the spatial arrangement of objects in an imageinvestigates the spatial arrangement of objects in an image4. Local orientation of the objects;4. Local orientation of the objects;6. Local orientation associated with the spatial arrangement of 6. Local orientation associated with the spatial arrangement of

similar objects.similar objects.Different Form

of Radical

Stroke

Different Formof Stroke

Page 18: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

David Marr's “Vision”David Marr's “Vision”investigates the spatial arrangement of objects in an imageinvestigates the spatial arrangement of objects in an image4. Local orientation of the objects;4. Local orientation of the objects;6. Local orientation associated with the spatial arrangement of 6. Local orientation associated with the spatial arrangement of

similar objects.similar objects.

Character

Radical

Different Formof Radical

Page 19: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

David Marr's “Vision”David Marr's “Vision”investigates the spatial arrangement of objects in an imageinvestigates the spatial arrangement of objects in an image5. Local distances associated with the spatial arrangement of 5. Local distances associated with the spatial arrangement of

similar objects;similar objects;

Different Formof Radical

Stroke

Different Formof Stroke

Page 20: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

David Marr's “Vision”David Marr's “Vision”investigates the spatial arrangement of objects in an imageinvestigates the spatial arrangement of objects in an image5. Local distances associated with the spatial arrangement of 5. Local distances associated with the spatial arrangement of

similar objects;similar objects;

Character

Radical

Different Formof Radical

Page 21: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

Page 22: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Automatic Generation Algorithm of Chinese Character

Character

Radical

Different Formof Radical

Stroke

Different Formof Stroke

Step 1: Input prior information of target character and absent radicals.Step 1: Input prior information of target character and absent radicals.Step 2: If all radicals to compose the character exist, then turn to Step 4.Step 2: If all radicals to compose the character exist, then turn to Step 4.Step 3: Compose the absent radicals using the Bayesian dynamic models, Step 3: Compose the absent radicals using the Bayesian dynamic models,

similar to (11)~(14).similar to (11)~(14).Step 4: Compose the target character using the Bayesian dynamic models.Step 4: Compose the target character using the Bayesian dynamic models.Step 5: Output the target character.Step 5: Output the target character.

Page 23: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Experiments and Discussion

Page 24: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Experiments and Discussion

Page 25: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

Experiments and Discussion

Page 26: Generation of Chinese Character Based on Human Vision and Prior Knowledge of Calligraphy

谢 谢 !谢 谢 !

http://www.icst.pku.edu.cn http://www.icst.pku.edu.cn