SEARCH

SEARCH BY CITATION

Keywords:

  • I.3.8 [Computer Graphics]: Applications;
  • I.3.3 Picture/Image GenerationDisplay algorithms

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. System Architecture
  5. 3. Muti-level Units Segmentation
  6. 4. Domain Ontology of Calligraphy Synthesis
  7. 5. Calligraphy Synthesis Based on Ontology
  8. 6. Experiments and Results
  9. 7. Conclusion
  10. 8. Acknowledgements
  11. References

There are lots of digitized Chinese calligraphic works written by ancient calligraphists in CADAL. Sometimes users want to generate a tablet or a piece of calligraphic work written by these calligraphists. However, some characters had not been written by them or though were ever written but are hard to read because of long time weathering. Furthermore, different works of the same calligraphist may have diverse styles. It is a great challenge to generate a new calligraphic character with particular style. In this paper, we present a novel approach to synthesize regular script calligraphic characters in the style of some specific calligraphic works. Our approach first processes original calligraphic resources by multi-level segmentation. Then, a domain ontology is established to supervise basic resources and to provide significant support for calligraphy synthesis. Finally, a novel algorithm is proposed to synthesize new calligraphic characters. We demonstrate the effectiveness of our method by showing various synthesized results in the style of different calligraphic works of Liu Gongquan.


1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. System Architecture
  5. 3. Muti-level Units Segmentation
  6. 4. Domain Ontology of Calligraphy Synthesis
  7. 5. Calligraphy Synthesis Based on Ontology
  8. 6. Experiments and Results
  9. 7. Conclusion
  10. 8. Acknowledgements
  11. References

Chinese calligraphy is an important traditional art form in China, which has more than 3000 years history.

image

Figure 1. Famous calligraphic works.

Download figure to PowerPoint

In CADAL (China Academic Digital Associative Library) calligraphic system1, 4136 famous ancient calligraphic works are digitized. With these calligraphic resources, we provide not only some basic functions like browsing and metadata-based searching but also some other functions like shape-based retrieval [ZZWL04], special style calligraphy rendering [ZWY10] and tablet generating [YWZ09].

Most calligraphic works were carved on stones or written on bamboos, silks or papers, which are fragile, hard to be reproduced and easily damaged by fire, flood and other reasons. Very few perfectly preserved calligraphic works of famous calligraphists have been handed down.

Several approaches have been proposed to address calligraphic character synthesis problem. In 1980s, Strassmann presented a 2-D virtual brush model [Str86]. In 2000, a 3-D virtual brush model was proposed [WI00]. In [LWI08] [WLI08] [WLI06], parameters of virtual brush model are extracted from a calligraphic character and used to reproduce calligraphic characters. Xu [XLCP05] used a six-level hierarchical model to represent Chinese characters and proposed a synthesis approach. An intelligent system for Chinese calligraphy [XJLP07] was proposed which could derive parameters in an interactive way and judge the beauty of calligraphy. However, the virtual brush model [Str86] [WI00] [LWI08] [WLI08] [WLI06] could only produce calligraphic characters which had been written by the calligraphists. Methods in [XLCP05] [XJLP07] could produce new calligraphic characters but the styles of them were not consistent. Yu [YWZ09] proposed a method for style-consistency calligraphy synthesis, which assumed the styles of all the works written by the same calligraphist are consistent, but in fact the styles of different works written by the same calligraphist may be different because of the change of time, place or mood. In addition, due to the difficulty in radical segmentation, most of the radicals used to synthesize characters in [YWZ09] were obtained by combining strokes, however, the original radicals can preserve the styles of the works much better than the synthesized radicals.

In this paper, we introduce a multi-level segmentation scheme to segment calligraphic resources and to get the multi-level units (characters, radicals and strokes). A domain ontology is established to supervise basic resources and to provide significant support for Chinese calligraphy synthesis. A novel algorithm is proposed to synthesize regular script calligraphic characters in the style of specific calligraphic works basing on the domain ontology.

The remainder is organized as follows. The system architecture is given in Section 2, multi-level segmentation is described in Section 3, establishment of domain ontology is in Section 4, Section 5 describes the ontology-based synthesis method in detail, the experimental results are presented in Section 6, in Section 7 a conclusion is given.

2. System Architecture

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. System Architecture
  5. 3. Muti-level Units Segmentation
  6. 4. Domain Ontology of Calligraphy Synthesis
  7. 5. Calligraphy Synthesis Based on Ontology
  8. 6. Experiments and Results
  9. 7. Conclusion
  10. 8. Acknowledgements
  11. References

As shown in Figure 2, the calligraphy synthesis system consists of two parts: The domain ontology of calligraphy synthesis and the synthesis factory.

image

Figure 2. The system architecture.

Download figure to PowerPoint

The domain ontology of calligraphy synthesis is composed of three sub-modules: calligraphic resource module (CRM), Chinese index module (CIM) and standard contour module (SCM). Since the calligraphic resources in CADAL are all in the form of calligraphic works, multi-level segmentation is introduced to decompose these works. The CRM is built basing on the multi-level calligraphic units (characters, radicals and strokes), which is used to provide materials for calligraphy synthesis. The CIM is established on the basis of GB23122, which is the most popular simplified Chinese character set with 6,763 characters and covers 99.99% commonly used characters. Composition and structure information of Chinese characters in GB2312 are collected and used in the CIM, which are significant and assistant knowledge for calligraphy synthesis. The SCM is built basing on a regular script TrueType3 font, which contains stroke contour information of all the characters in GB2312. This module is not only important for calligraphic radical synthesis but also a basic reference for calligraphic character synthesis.

The synthesis factory includes two workshops: the calligraphic radical workshop and the calligraphic character workshop. In the radical workshop, calligraphic radicals are synthesized by calligraphic strokes selected from CRM with the aid of radical contours in SCM. In the character workshop, calligraphic characters are generated by calligraphic radicals, which are either selected from CRM or synthesized by the radical workshop.

The process of calligraphy synthesis is also shown in Figure 2. To get a calligraphic character in the style of a specific calligraphic work, Chinese character and the constrained conditions are inputted firstly. If the corresponding calligraphic characters exist in the CRM, the one with the highest priority is outputted directly, as the red lines show. While if not exist, CIM is looked up to determine radical composition of the Chinese character and appropriate calligraphic radicals are selected from CRM to synthesize new calligraphic character, as the green lines show. However, if the radical is not found, stroke composition of the radical is obtained from CIM and appropriate calligraphic strokes are chosen from CRM referring to stroke contours in SCM. Calligraphic radical is synthesized by combining these calligraphic strokes and then new calligraphic character is generated in the character workshop, as the blue lines show.

3. Muti-level Units Segmentation

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. System Architecture
  5. 3. Muti-level Units Segmentation
  6. 4. Domain Ontology of Calligraphy Synthesis
  7. 5. Calligraphy Synthesis Based on Ontology
  8. 6. Experiments and Results
  9. 7. Conclusion
  10. 8. Acknowledgements
  11. References

As mentioned above, most of the calligraphic resources in CADAL are in the form of works. The works should be decomposed into different granularity units for further use. Minimum bounding box method is used to separate individual calligraphic characters from calligraphic works and an interactive polygon-based approach is applied for calligraphic radicals and strokes segmentation.

3.1. Minimum Bounding Box Method

Individual calligraphic characters, the top level units, can be separated from calligraphic works by minimum bounding box method [MS99]. First of all, binaryzation and noise reduction are applied to preprocess the calligraphic works. Since ancient calligraphic works were always written longitudinally, as shown in Figure 3, black pixels in the image are mapped to the x-axis first, single column characters are segmented according to the distribution of black pixels on the x-axis. Then black pixels in the single column are mapped to the y-axis, individual characters are segmented depending on the distribution of black pixels on the y-axis.

image

Figure 3. Character segmentation.

Download figure to PowerPoint

3.2. Polygon-based Method

Due to the complex position relationship between radicals and between strokes, as shown in the upper part of Figure 4, it is difficult to separate radicals from characters or to extract strokes from radicals only by minimum bounding boxes. So, an interactive polygon-based unit segmentation (IPUS) approach is proposed to obtain radicals and strokes. To get radicals from calligraphic character images, as shown in the lower part of Figure 4, firstly, a polygon is drawn surrounding the radical to locate the position. A pixel points set P = {(x0, y0), (x1, y1),…, (xn–1, yn–1)} is used to represent the polygon. The detailed process of IPUS-based radical segmentation is shown in Algorithm 1, Q is a FIFO queue of points and Visited is a boolean matrix to record whether points in the original image have been visited.

image

Figure 4. Radical segmentation.

Download figure to PowerPoint

Algorithm 1: Algorithm of IPUS.

Image

At last, the marginal blank area of RadicalImg is removed and position information of the radical in the original image is updated and stored in the CRM. IPUS is also suitable for calligraphic strokes extraction, but unlike system font, calligraphic works are quite vague and have lots of noise because of years of weathering, which make strokes extraction inconvenient. It will take a lot of manpowers and time to extract all strokes from radicals instances in CRM. Although the stroke styles of calligraphy are diversified and strokes may be expressed in different shapes, a majority of calligraphic characters are composed with some basic calligraphic strokes with little transformation. Therefore, in the stroke part of CRM, calligraphic strokes are classified more refinedly than Chinese basic strokes. A certain amount of samples belonging to each kind of calligraphic stroke are extracted and a calligraphic stroke library is built for each calligraphic work which will be used to synthesize calligraphic radicals.

4. Domain Ontology of Calligraphy Synthesis

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. System Architecture
  5. 3. Muti-level Units Segmentation
  6. 4. Domain Ontology of Calligraphy Synthesis
  7. 5. Calligraphy Synthesis Based on Ontology
  8. 6. Experiments and Results
  9. 7. Conclusion
  10. 8. Acknowledgements
  11. References

In order to conveniently organize resources and to establish the connection between different parts of resources, a domain ontology of calligraphy synthesis is built, which is the foundation of style consistent calligraphy synthesis. The ontology not only makes the data structure clearer, but also has good scalability, which means that a new style can be easily added to synthesize new calligraphic characters. In the description below, label is the semantic meaning of the Chinese character or the Chinese radical.

4.1. Analysis of Chinese Character

Chinese characters are the basis of calligraphic characters and each calligraphic character corresponds to a Chinese character. Therefore, it is necessary to analyze Chinese characters firstly to find the composition and structure features of Chinese characters. Level-model and block-model are proposed to represent the composition and structure of Chinese characters clearly. They are also the foundation of the calligraphy synthesis domain ontology establishment. All of the following sub-modules are set up on the basis of the two models.

Level-model is used to represent the composition of Chinese characters. As shown in Figure 5(a), a three level model, including characters, radicals and strokes, is built. The radical composition of character and stroke composition of radical are distinct in the level-model.

image

Figure 5. (a) level-model; (b) block-model.

Download figure to PowerPoint

Chinese characters can be divided into 13 basic structures according to the position of radicals, as shown in the table 1. Obviously, each Chinese character has up to three radicals and a block-model is proposed basing on the 13 basic structures. There are five boxes in the block-model, top, right, bottom, left and center. Since the amount of HS and FS is relatively few, it should be stated that when dealing with structures of FS and HS, we put the outside radicals into the left box artificially. The structure information of Chinese characters can be expressed using the block-model, as shown in Figure 5(b).

Table 1. Basic structures of Chinese character.
Image

4.2. Calligraphic Resource Module (CRM)

Calligraphic resource module mainly contains instances of calligraphic characters, calligraphic radicals and calligraphic strokes. As shown in Figure 6, a three-level graph is built for this module. The function of this module is to provide calligraphic materials for calligraphy synthesis. It should be noted that some attributes in this module can be inherited from the top level to the lower level, we call them i-attribute, however, some attributes can't and we call them n-attribute. Top-level: Calligraphic character

image

Figure 6. Calligraphic resource module.

Download figure to PowerPoint

       {label, author, work, priority}

Work is the name of the calligraphic work where the calligraphic unit comes from and author is the name of the calligraphist who wrote the calligraphic unit. Priority is the quality of the calligraphic character, which has three levels (0,1,2) and is obtained in the process of annotation. The attributes underlined are inherited.

Mid-level: Calligraphic radical

       {label, priority, block, position}

Priority is the quality of the calligraphic radical, block is the block which the radical belongs to in the block-model. Position is position information of the radical, which is obtained in the process of radical segmentation and represented by a 4-dimension vector:

       L = (topX, topY, bottomX, bottomY)

(topX, topY) and (bottomX, bottomY) are the left-top point and the right-bottom point of the radical in the original calligraphic character image. Bottom-level: Calligraphic stroke

       {type}

Type is the basic stroke type of the calligraphic stroke. Constraints: Each calligraphic character relates to n calligraphic radicals (1 ≤ n ≤ 3) and each calligraphic radical links to m calligraphic strokes (1 ≤ m). The calligraphic characters are independent each other, there do not exist the cases that a number of calligraphic character instances associate with one calligraphic radical instance or some calligraphic radical instances associate with one calligraphic stroke instance.

4.3. Chinese Index Module (CIM)

Chinese index module contains instances of Chinese characters, Chinese radicals and Chinese strokes. As shown in Figure 7, a tripartite graph is established to represent the association between instances. This module is mainly used to provide auxiliary information for calligraphy synthesis. In CIM, instances only contain n-attribute.

Top-level: Chinese character

image

Figure 7. Chinese index module.

Download figure to PowerPoint

           {label, structure, cluster}

Structure is the basic structure type of the Chinese character. Cluster is the classification result of the Chinese character, which will be described in 4.4.

Mid-level: Chinese radical

           {label, block, order}

Block means the block which the radical belongs to in the block-model. Order is the strokes order of the radical. Bottom-level: Chinese stroke

           {type}

Type is the basic stroke type of the Chinese stroke.

Constraints: Five basic types stroke are defined Image. Each character can associate with n radicals (1 ≤ n ≤ 3). Radicals with the same label but in different blocks are considered as different radicals and correspond to different instances. Each calligraphic radical is connected to multiple strokes. Radicals can be shared between characters and strokes can be shared between radicals. It is common that a radical may contain a plurality of same kind basic strokes, such as Image contains 3 Image. Therefore, numbers are marked on the links between radicals and strokes, which make the corresponding relationship clearer.

4.4. Standard Contour Module (SCM)

The standard contour module contains different levels of regular script contour instances. The TureType font of GB2312, which contains strokes contour information of all common characters, is used to assist calligraphic radicals synthesis and to provide a reference in the process of calligraphic characters synthesis. Contours are represented by straight lines and cubic bezier curves in the regular script font of TrueType, so it is easy to obtain contour information from SCM. This module is similar to calligraphic resource module, but it does not have i-attribute, as shown in Figure 8.

image

Figure 8. Standard contour module.

Download figure to PowerPoint

Top-level: Character contour

           {label}

Mid-level: Radical contour

           {label, position, block}

Block is the block which the radical belongs to in the block-model and position is position information of the radical contour.

Bottom-level: Stroke contour

           {type, position}

Type is the basic stroke type of the stroke and position is position information of the stroke contour.

Constraints: A character contour is associated with n radical contours (1 ≤ n ≤ 3), a contour of radical corresponds to m strokes contours (1 ≤ m). Character contours are independent and there are no public radical contours of the character contours, so are radical contours.

In [YWZ09], Chinese characters are classified by thirteen basic structures and when a calligraphic character is synthesized, the layout of radicals is determined according to the type of structure. However, as shown in Figure 9, although the four characters have the same type of structure, the layouts of radicals are obviously different. It is impractical to only use the same layout of radicals to synthesize characters with the same structure. Thus, a more refined method basing on both the structure and radical position information in SCM is used to classify Chinese characters in GB2312.

image

Figure 9. Characters with the same structure but they are different in the layout of radicals.

Download figure to PowerPoint

Here, a 4 × n matrix M = (L1, L2,…, Ln) is used to represent the radicals layout of a Chinese character, where n is the number of radical contours and Li(i = 1,…, n) denotes the radical position information. Since characters with the same structure type have the same amount of radicals, we can classify Chinese characters with different structures respectively. K-means is used to divide Chinese characters of each structure into k clusters, where k is empirical.

4.5. Modules Integration

The three sub-modules above are integrated to create domain ontology of calligraphy synthesis. Since the three modules are all built on the basis of level-block, we construct the connection between modules depending on hierarchy, the Chinese index module becomes the link between the modules. An example is shown in Figure 10.

image

Figure 10. Example of modules integration.

Download figure to PowerPoint

Top-level: Contact is established through the label, characters instances with the same label are linked together. The relationship between Chinese characters and calligraphic characters is one-to-many, however, one-to-one relationship is between Chinese characters and standard character contours.

Mid-level: Contact is established through the label and the block. A Chinese radical corresponds to multiple calligraphic radicals and also corresponds to multiple standard radical contours.

Bottom-level: Basic stroke types and strokes order are used to build the connection. One-to-many relationship is applied in both CIM to CRM and CIM to SCM in the level of stroke.

5. Calligraphy Synthesis Based on Ontology

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. System Architecture
  5. 3. Muti-level Units Segmentation
  6. 4. Domain Ontology of Calligraphy Synthesis
  7. 5. Calligraphy Synthesis Based on Ontology
  8. 6. Experiments and Results
  9. 7. Conclusion
  10. 8. Acknowledgements
  11. References

Calligraphic characters, especially regular script, in different calligraphic works are mainly different in style of strokes and layout of radicals. When a character ch in the style of a specific calligraphic work w written by au is needed, we should generate it. An algorithm based on ontology is proposed for calligraphic character synthesis. In the process of synthesis, there are three levels. On the first level, we search calligraphic characters in CRM to get corresponding calligraphic character instances, if found, output the one with the highest priority. However, if not found, go to the second level. On the second level CIM is searched to determine the radical composition of the Chinese character, appropriate calligraphic radical instances are selected from CRM to synthesize the calligraphic character. Nevertheless, if the calligraphic radical is not found, stroke composition of the radical is determined in CIM and calligraphic strokes are selected from CRM basing on the corresponding stroke contours in SCM, then the calligraphic radical is generated. In the end, the calligraphic character is synthesized by using calligraphic radicals in CRM or from the radical workshop. The detailed algorithm is shown in Algorithm 2 and the detailed processes of stroke selection, radical synthesis, radical selection and character synthesis are described below.

Algorithm 2. Algorithm of calligraphy synthesis

Image

5.1. Radical Selection

The styles of different calligraphic works are diverse. It is better to select radicals with similar style of the specific calligraphic work to synthesize new characters. Since different calligraphic works were written by different kind of writing brushes, which decide the average stroke width of the works. In addition, although stroke width of individual calligraphic stroke is various and contributes to the aesthetic of calligraphy, average stroke width of characters in the same work tend to be similar. Thus, stroke width is a useful reference for radical selection. The radical with the highest matching degree is selected.

The matching degree of radical x is defined:

  • display math(1)

where W(x) and P(x) are indicator functions:

  • display math(2)
  • display math(3)

Q(x) is the quality of radical x, SW(x) is the average stroke width of radical x and sw is the average stroke width of all the calligraphic characters in the calligraphic work w. α, β, η and ω are empirical weighting coefficients. The SW(x) is expressed:

  • display math(4)

where area(x) is the number of pixels in the binarized image of radical x, contourLength(x) is the number of pixels in the contour image of radical x. If the best calligraphic radical is from other works, erosion and dilation in mathematical morphology are applied to keep average stroke width of the radical consist with sw.

5.2. Stroke Selection

If the radical is not found or the matching degrees of the radicals are below a threshold θ, appropriate strokes are to be selected from the calligraphic stroke library of the specific work in CRM to synthesize new calligraphic radical. Firstly, the corresponding radical instance in CIM is searched to determine the stroke composition of the radical and then appropriate calligraphic strokes are chosen by comparing with the corresponding standard stroke contours in SCM. Calligraphic stroke images are scaled to the same size of the stroke contours and the one with the highest similarity degree is selected to synthesize new radical.

The similarity degree of stroke x is defined:

  • display math(5)

where w and h are the width and height of the calligraphic stroke x, λ1 and λ2 are empirical weighting coefficients, fs and gs are critical functions of stroke s:

  • display math(6)
  • display math(7)

5.3. Radical Synthesis

Since the relative position relationship of strokes in regular script is basically fixed, position information of strokes in TrueType font is used and the radical is synthesized by the strokes selected in 5.2. as shown in Figure 11. Firstly position information of strokes is obtained from SCM. Then calligraphic strokes are scaled to match the position information and finally the radical is synthesized.

image

Figure 11. Process of radical synthesis.

Download figure to PowerPoint

5.4. Character Synthesis

Here the calligraphic character image chImg is generated by calligraphic radicals. Obviously, the radicals layout of the character is unknown because the calligraphic character we need to synthesize did not exist in the calligraphic work w. Since structure of regular script font is relatively stable, radicals layouts of TrueType font in SCM are perfect references for character synthesis. However, different calligraphic works have their unique and relatively fixed style, so radicals layouts of other calligraphic characters in w are analyzed and used to indicate the unique style of w. Three factors are considered to determine the radicals layout. (1) C is the corresponding character instance in SCM. (2) Result of Chinese character classification is applied. P = {C1, C2, Cn } is a set of calligraphic characters from w and in the same cluster of ch. (3) Same radicals in different calligraphic characters almost have very similar position information. Q = {C1, C2, Cm} is a set of calligraphic characters from w, which have the same radical with ch. The position information L of radical φ in chImg is defined as follows:

  • display math(8)
  • display math(9)

where χ1, χ2 and χ31 + χ2 + χ3 = 1) are empirical weighting coefficients to adjust the radicals layout. It should be noted that if P is empty, set χ2 = 0 and if Q is empty, set χ3 = 0. FS is the font size of instances in SCM. As shown in Figure 12, calligraphic character is synthesized by radicals.

image

Figure 12. Result of character synthesis.

Download figure to PowerPoint

6. Experiments and Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. System Architecture
  5. 3. Muti-level Units Segmentation
  6. 4. Domain Ontology of Calligraphy Synthesis
  7. 5. Calligraphy Synthesis Based on Ontology
  8. 6. Experiments and Results
  9. 7. Conclusion
  10. 8. Acknowledgements
  11. References

Three different regular script calligraphic works, XuanMiTaBei(XMTB), GuiLinShiBei(GLSB) and Tai-HeTie(THT) of Liu Gongquan (a famous calligraphist in Tang Dynasty) are chosen to be the style templates of synthesis in our experiments.

Four kinds of experiments are executed to verify the effectiveness of the ontology-based calligraphy synthesis approach: (1) The detailed synthesizing process of a calligraphic character image; (2) Style comparison of synthesized characters and the original ones; (3) Style comparison of synthesized results with styles of different calligraphic works; (4) Evaluation of the synthesized results. In these experiments, the original character instances are deleted from CRM to guarantee that the synthesized results will not be influenced by them.

Parameters are set: α = 0.4, β = 0.1, e = 0.4, ω = 0.1, θ = 0.6, λ1 = λ2 = 0.5, χ1 = 0.5, χ2 = 0.3, χ3 = 0.2.

6.1. Synthesis of a Calligraphic Character

In this experiment, the whole process of synthesizing a calligraphic character “Image” in the style of XuanMiTaBei is described. The character, calligraphist and calligraphic work are inputted and as shown in Figure 13, CIM is searched to determine the radical composition of the character.

image

Figure 13. Radical composition of the character.

Download figure to PowerPoint

The left radical is found in CRM and the searched results are shown in Figure 14. The matching degree of each “Image” radical is calculated by formula (1) and the calligraphic radical with the highest matching degree is chosen for further synthesis.

image

Figure 14. Searched results of the left radical in CRM. The characters above the blue line are from w and characters below the blue line are from other works. The radical surrounded by red box is selected for synthesis

Download figure to PowerPoint

Unfortunately, the right radical is not found in CRM, so it should be synthesized by the calligraphic strokes in the stroke library of the specific calligraphic work. As shown in Table 2, CIM is searched to determine the stroke composition of the radical, calligraphic strokes are selected from CRM by comparing with the corresponding stroke contours in SCM. At last, calligraphic radical is synthesized by combining the calligraphic strokes depending on position information of stroke contours.

Table 2. Synthesis process of calligraphic radical.
Image

As shown in Table 3, radicals layout of the new character is got by adjusting the three factors described in 5.4. The result of synthesis is shown in Figure 15.

Table 3. Radical Layout of the character.
Radical layoutFactor1Factor2Factor3Result
LR.topX0110
LR.topY0162
LR.bottomX841087887
LR.bottomY173198196187
RR.topX84106094
RR.topY3654044
RR.bottomX1971980197
RR.bottomY1581720164
LR is short for left radical and RR is short for right radical. Since the right radical is not found in CRM, position information of the right radical in factor 3 is all 0. In this case we set χ1 = χ2 = 0.5.
image

Figure 15. Result of the synthesis.

Download figure to PowerPoint

6.2. Style Comparison of Synthesized Results and the Original Ones

In this experiment, as shown in Table 4, synthesized calligraphic character images are compared with the original ones. Furthermore, in order to make the observation convenient, all calligraphic characters are adjusted to the same size.

Table 4. Style Comparison of Synthesized Results and the Original Ones.
Image

6.3. Synthesized Results of Different Calligraphic Works

In this part, a Chinese idiom, which means the beauty of the spring, is synthesized in the style of three different calligraphic works. As shown in Figure 16, different works express different styles.

image

Figure 16. Style comparison of different calligraphic works.

Download figure to PowerPoint

6.4. Identification of the Synthesized Results

In this part, 300 characters are synthesized by the synthesis system, as shown in Figure 17.

image

Figure 17. Results of calligraphic character synthesis.

Download figure to PowerPoint

Some terrible synthesized characters are analyzed to find the reasons that lead to the bad results. After analysis, the main reasons are summarized as follows:

  1. Styles of radicals are inconsistent. Radicals from other calligraphic works are used to synthesize calligraphic characters, which make styles of radicals inconsistent. Although strokes width of radicals from other works is adjusted to keep style consistent, we can see the difference of style among radicals (As shown in Figure 18(a)(b)).
    image

    Figure 18. Terrible results of synthesis.

    Download figure to PowerPoint

  2. Position of radicals is irrational. The distance between radicals sometimes is too far, making the two radicals look like two single structure characters, as shown in Figure 18(c), but sometimes the distance is too close, making the radicals cross each other, as shown in Figure 18(d).
  3. Inappropriate calligraphic strokes. Inappropriate calligraphic strokes are chosen to synthesize calligraphic radicals. As shown in Figure 18(e)(f), the strokes in red box are inappropriate and make the results terrible. Since the shapes of “zhe” (a kind of basic stroke) are diverse, inappropriate strokes always appear when selecting this kind of strokes.

7. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. System Architecture
  5. 3. Muti-level Units Segmentation
  6. 4. Domain Ontology of Calligraphy Synthesis
  7. 5. Calligraphy Synthesis Based on Ontology
  8. 6. Experiments and Results
  9. 7. Conclusion
  10. 8. Acknowledgements
  11. References

In this paper, an ontology-based approach is proposed to synthesize special style calligraphic characters of regular script. A multi-level segmentation scheme is used to obtain different levels of calligraphic resources. A domain ontology is established to supervise resources and to provide significant support for calligraphy synthesis. Then a novel algorithm is proposed to synthesize calligraphic characters with specific styles. In our experiments, characters are synthesized in the style of three calligraphic works of Liu Gongquan and the results show that our proposed methods are effective.

8. Acknowledgements

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. System Architecture
  5. 3. Muti-level Units Segmentation
  6. 4. Domain Ontology of Calligraphy Synthesis
  7. 5. Calligraphy Synthesis Based on Ontology
  8. 6. Experiments and Results
  9. 7. Conclusion
  10. 8. Acknowledgements
  11. References

This work is supported by National Natural Science Foundation of China (No. 61070066), the CADAL Project and Research Center, Zhejiang University.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. System Architecture
  5. 3. Muti-level Units Segmentation
  6. 4. Domain Ontology of Calligraphy Synthesis
  7. 5. Calligraphy Synthesis Based on Ontology
  8. 6. Experiments and Results
  9. 7. Conclusion
  10. 8. Acknowledgements
  11. References
  • 1
    [LWI08] Leung H., Wong S. T., Ip. H. H.: In the name of art[j]. IEEE Signal Processing Magazine, 25 (2008), 4954. 1.
  • 2
    [MS99] Manmatha R., Srimal. N.: Scale space technique for character segmentation in hand-written manuscripts. International Conference on Scale-Space Theories in Computer Vision (1999), 2233. 3.
  • 3
    [Str86] Strassmann S.: Hairy brushes. ACM Siggraph (1986), 225232. 1.
  • 4
    [WI00] Wong H. T. F., Ip H. H. S.: Virtual brush: A model-based synthesis of chinese calligraphy. Computer Graphics 24 (2000), 99113. 1.
  • 5
    [WLI06] Wong S. T. S., Leung H., Ip. H. H. S.: Skeleton-based analysis of oriental calligraphic images. Pattern Recognition and Machine Intelligence 1 (2006), 121. 1.
  • 6
    [WLI08] Wong S. T., Leung H., Ip H. H.: Model-based analysis of chinese calligraphy images. Computer Vision and Image Understanding 109 (Jan. 2008), 6985. 1.
  • 7
    [XJLP07] Xu S., Jiang H., Lau F. C., Pan. Y.: An intelligent systemfor chinese calligraphy. Conference on Association for the Advancement of Artificial Intelligence (Jul. 2007), 15781583. 1.
  • 8
    [XLCP05] Xu S., Lau F. C., Cheung W. K., Pan Y.: Automatic generation of artistic chinese calligraphy. IEEE Intelligent Systems (May. 2005), 3239. 1.
  • 9
    [YWZ09] Yu K., Wu J., Zhuang Y.: Style-consistency calligraphy synthesis system in digital library. Joint Conference on Digital Libraries (Jun. 2009), 145152. 1, 2, 5.
  • 10
    [ZWY10] Zhang Z., Wu J., Yu K.: Chinese calligraphy specific style rendering system. Joint Conference on Digital Libraries (Jun. 2010), 100108. 1.
  • 11
    [ZZWL04] Zhuang Y., Zhang X., Wu J., Lu X.: Retrieval of chinese calligraphic character image. Pacific-Rim Conference on Multimedia (Nov. 2004), 1714. 1.