CHINESE CALLIGRAPHY IMAGE ANALYSIS PROJECT

Stroke Extraction for Chinese Calligraphy Images

CONTENT

• Team Members
• Introduction
• Prior Work
• Objectives
• Project Description
• Publication
• Contact

TEAM MEMBERS

• Howard Leung  (howard@cityu.edu.hk)
• Hui Kang  (huikang@cityu.edu.hk)

Top of this page

INTRODUCTION

Chinese calligraphy is the most fundamental artistic manifestation for Chinese people. Calligraphy characters are not used only for presenting the meaning to readers, but can also be appreciated as a form of art through various combinations of strokes with certain degree of flexibility. Nowadays, with the advance of technology, Chinese calligraphic artwork can be scanned and stored in the form of images. Given such a character image, the objective of stroke extraction is to extract the individual strokes that form the character. In this project, we propose an algorithm to extract the individual strokes of a Chinese calligraphy character to preserve the original shape of the calligraphy as closely as possible. We perform experiments on three most common styles of calligraphy, Kai-style, Cao-style and Li-style.

Top of this page

PRIOR WORK

Stroke extraction has been considered as a critical part of Optical Character Recognition (OCR) based on structural analysis. Many methods for off-line Chinese character recognition have been proposed to handle printed text font or handwritten characters. As Chinese calligraphy is more complicated and has more variations compared with printed font or handwritten character, the prior work of stroke extraction does not perform well when applied for Chinese calligraphy character images.

Top of this page

OBJECTIVE

• Extract the strokes forming the character from Chinese calligraphy images
• Merge the stroke segments accurately to form complete strokes

Top of this page

PROJECT DESCRIPTION

1. Preprocessing: The calligraphy character image is preprocessed by a thinning algorithm. Feature points, i.e. fork point and end point are detected.

2. Fork point grouping: Because of thinning distortion, one fork point is sometimes split into two fork points. Therefore a fork point grouping criterion should be used to get the correct fork point. We propose a new fork point grouping criterion that can achieve a better result, compared with the existing Maximum Circle Criterion approach.

3. Polygonal Singular Region: Here a novel concept of Polygonal Singular Region proposed for dividing a character into several stroke segments and preserving the original shape of the stroke more closely.

Stroke Merging: After the Polygonal Singular Region divides a connected region into stroke segments, we should decide that whether and which two stroke segments should be merged. It is not a trivial problem as Chinese writing is complicated and full of variations, let alone the characters written in a calligraphy style. We propose a stroke merging method based on the direction and the smoothness of the boundary points of stroke segments.

5. Experiment Results

Kai-style:

         
          
Cao-style:

          

Li-style:

         

Top of this page

PUBLICATION

•  Howard Leung and Hui Kang, "Stroke Extraction for Chinese Calligraphy Images", Asia Pacific Workshop on Visual Information Processing (VIP2005), Hong Kong, December 2005.

CONTACT

Any suggestions or comments are welcome. Please send them to Howard Leung. 

Top of this page