CHAPTER 1 INTRODUCTION
1.3 Thesis Organization
This thesis is organized as follows. In chapter 2, we review the technique of computer-synthesized Chinese-Ink Painting in Non-Photorealistic Rendering. Then, we briefly introduce the features of Chinese Fine-Brushwork Painting in chapter 3. In chapter 4, we introduce the stroke model and the stroke generating procedure in detail.
Chapter 5 describes the procedure for color diffusion. Then our implementation results are shown in chapter 6. Finally, chapter 7 is the conclusions and future works.
Figure 1.1 The system flowchart
Chapter 2
Related works
In this chapter, we discuss previous researches related to Chinese Ink Painting.
We focus on two topics: the brush model, and the ink and color diffusion. Then, we also talk about the works which render an input image in non-photorealistic style.
2.1 Brush Model
In order to simulate the characteristics of real brushes, the researcher often needs to define an appropriate brush model. Strassmann [17] provided a fundamental model on this topic. He applies the physical properties of the real brush material in his model and well simulates the traditional Japanese art of sumi-e. His brush model is described by a one-dimensional array which sweeps along a spline curve. Suguru Saito and Masayuki Nakajima [15] also established a brush model for Japanese calligraphy and sumi-e paintings. But, it is a three-dimensional brush model which let the users can draw the strokes by using a pen-style input device. The same as Saito’s work, the DAB [5] project proposed an actual pen-style input device for users. It is more flexible in terms of brush shape, dynamics, and loading. It also can be implemented by three-dimensional graphics hardware.
The above techniques are for Western paintings or Japanese paintings. There are also several researches [7,19,20,21] simulating Chinese Writing Brush in Chinese Ink Painting. Shan-Zan Weng [19] introduced a two-dimensional brush model. He defined a variable circular area as the contact region of brush and canvas. In this area, the bristles are distributed uniformly. The center of the circle will move along the spline and sweep the footprints as the ink calligraphy on the canvas. He provided several types of brush, so the user can generate different kinds of stroke by selecting proper parameters. Wong and Ip [20] used an inverse cone to represent the virtual brush.
They defined some interrelated parameters to change the density, opacity, and the shape of a footprint which corresponds to different behaviors of the brush.
According to the feature of Chinese Fine-Brushwork Painting, we define the brush model similar to Weng’s work in our system. Its advantages are easy implementation and well performance.
2.2 Ink and Color Diffusion
The effect of ink diffusion produced by the incredible absorbency of Hsuan paper is the most important feature of Chinese Ink Painting. So, several studies focused on this topic. [11,12,22] The technique proposed by Lee [12] rendered black ink paintings efficiently with realistic diffusion effects. It can simulate different kinds of diffusion depends on the kinds of paper or ink properties. But, it can not handle the blending effect of two or more strokes. Sheng-Wen Huang et al. [11] presented a method for simulating diffusion based on physical theory and analysis of observations.
Their study discussed the relationship with water particles, carbon particles and Hsuan papers when the ink diffusing. They also contributed the expression of a mixture of
different kinds of brush strokes, such as those of two wet brushes.
About color diffusion, in the DBA project [5], they used a simple alpha blending to show the color mixture. Curtis and Anderson [8] used Kubelka-Munk theory to simulate the effect of it. This theory studies the light reflection and transmission between pigments and paper. Another work that used KM theory is proposed by Wei-Jin Lin [13]. He employed KM model for the blending of traditional Chinese colors.
In this thesis, we use a new technique for synthesizing ink and color diffusion based on actual painting procedure and analysis of observations. Unlike the previous works of Chinese Ink Painting, we simulate the results of drawing on the alum Hsuan paper（礬宣）which is the particular material for Chinese Fine-Brushwork Painting.
2.3 Non-photorealistic Rendering
Researches in non-photorealistic rendering can be classified into two categories.
The first one focuses on simulating and modeling the artist’s brushes. They often supply an interactive interface such as Painter [5,7,8,13,15,22] for users to create their compositions. The researches in this topic are described by section 2.1 and 2.2. The other [9,10,14,18,19] usually takes an image as the input and applies user-defined pattern or some filters of image processing in order to transform the input image into NPR styles.
Haeberli [9] proposed a stroke-based rendering technique with parameters control. These parameters describe properties of a stroke, such as color, length, width,
and the location. Based on this technique, several researches are proposed to synthesize different NPR styles, for instance, a style of pencil drawing [14], oil painting [10] or Chinese ink painting [18,19]. Shan-Zan Weng [19] presented a system which inputs an existing Chinese Ink Painting image and re-paints it in other Chinese Painting styles. The system extracts the skeleton of the stroke automatically and applies a brush model with different style to it. In Chain-Ru Tang’s [18] work, it transformed an animal image into Chinese Fine-Brushwork style. It generates the hair drawing by using radial basis functions (RBFs) and makes use of mean shift filter to synthesize color diffusion.
In this thesis, we construct a system for the categories depicted before. Similar to Weng’s work, we take one image as the input and extract the skeletons and the colors on it. Then, we apply a brush model and color diffusion according to this information and then transform the image into Chinese Fine-Brushwork style.
Chapter 3
Preliminary of Chinese Fine-Brushwork Painting
3.1 An Overview
Chinese Fine-Brushwork Painting has a long history. It began to develop from Epoch of Warring States until the Tang dynasty did it became mature enough to compare with the Free Style Chinese Ink Painting. However, during the Tang dynasty the Fine-brushwork Painting on Figure prevailed, a famous artist Zhou Fang’s work
“Ladies wearing flowers”（周昉 『簪花仕女圖』） which depicted the real life with strong atmosphere of poetry. At that time the Birds and Flowers Painting was still in the initial stage. During the Five Dynasties, Huang Quan （黃荃） founded the groundwork of the Birds and Flowers Painting with sketching the contour doubly and coloring （”雙勾填彩”）.
And he also created the method to dye with multi-layer which means to dye repeatedly on the same location. （”三暈九染”） These techniques are still influential nowadays. Not until the Sung dynasty did Birds and Flowers Painting became mature and prosperous. Gorgeous and fine work was its characteristic, so it was taken seriously by the royal court such as Li Sung’s
work “a flower basket” （李嵩 『花籃圖』）. Between the Yuan and Ming Dynasties, the Literati Painting became the main stream. So it procured the decline of Fine-Brushwork Painting and the popularity of Free Style Painting on Birds and Flowers. But, it found a new way in the Ching dynasty. Influenced by the culture of Western, the Fine-Brushwork Painting developed a new style. It became more realistic, three-dimensional and vivid, because of the entering of Western Painting techniques, for example, the Lang Shining’s work “the Flowers”.
（朗世寧 『聚瑞圖』）[2]
Figure 3.1 Creations of Fine-Brushwork Painting
On the whole, Fine-Brushwork Painting emphasizes the reality of objects. Artists always work carefully and neatly with various colors. Therefore, they recently choose Hong-do pen （紅豆筆） which is composed of firm wolves’ hair. This kind of pen is quite suitable to draw delicate and flexible lines. They tend to choose pen with soft hair to dye with the technique of multi-layer. In the respect of using paper, paper of
(a)“A flower basket” by Li Sung
(b) “The flowers”
by Lang Shining [2]
Fine-Brushwork Painting is much different from that of Free Style painting. It adopts alum-Hsuan （礬宣） paper. It is one kind of Hsuan paper on which we brush a layer of alum. Because alum has immobility, alum-Hsuan paper does not be permeated by water easily and it does not make the ink diffuse. Therefore, it is quite suitable for Fine-brushwork Painting with several times of dyeing.[3,4]
In this thesis, we focus on the equipments’ characteristics of the Fine-Brushwork Painting which are the Hong-do pen and the alum-Hsuan paper.
3.2 The Painting Procedure
In this section, we explain the procedure of drawing Fine-Brushwork Painting on flowers. We use the compositions created by Liu Yu Hsia （劉玉霞） who has more then ten-year experiences in this work to expound it .[4]
Figure 3.2 The work of Liu Yu Hsia
The process of Fine-Brushwork Painting is divided into two types: sketching the
contour and coloring.[1,2] First, painters draw the outline of the flower by using pale ink. It is the draft of the composition. Next, they use washing technique （”渲染”）
which is a traditional skill of Chinese Painting to dye the objects. For washing, painters use two brushes that one is soaked in water and the other is soaked in colors.
Painters employ the colored pen to dye the color inside the lines. Then, they extend it inside out by using the watered pen. It causes the color decreasing bit by bit from inner to outer and makes the color vary plentifully. In the Fine-Brushwork Painting, washing is the major skill for coloring. Painters use it to dye layer by layer and let the creations colorful. For instance, in flower coloring, painters usually apply it two or more times with different colors. Painters may apply it from the petals to the pistils with white color first, then use it again inversely with the color depending on the flowers. Finally, they sketch the contours again and draw the detail of the objects.
Figure 3.3 shows this process.
Figure 3.3 A procedure of painting
(a) Sketch the contour (b) First washing
(c) Second washing (d) Draw the detail
Chapter 4
Stroke Generation
In this chapter, we describe how to obtain the stroke information from input image and use it to sketch the contour in the style of Chinese Fine-Brushwork Painting. We introduce our proposed brush model first, and then discuss it from three parts: the petal, the leaf and the pistil. Figure 4.1 shows the flowchart of sketching.
In Chinese Fine-Brushwork Painting, sketching the contour is the first step of all.
First, Painters draw the shapes and positions of objects realistically by using pale ink.
It is the draft for dyeing. Then, after dyeing, they sketch the contours again by using thick ink or any other preferred color. It may make the creations more vivid. The Hong-do pen（紅豆筆）is the equipment in common use for this procedure. So we simulate our brush model for the Hong-do pen style.
This chapter is organized as follow. In section 4.1, we talk about our proposed brush model that is based on the Hong-do pen style. Then, we describe the part of flower in section 4.2 that contains the edge extraction, the stroke definition, and the brush model execution. Section 4.3 explains the generation of veins. In the section 4.4, we introduce how the pistil drawing and its color determination.
Figure 4.1 The flowchart of sketching the contour
4.1 The Brush Model of Hon-Do Pen
In the field of Chinese Fine-Brushwork, artists use tapering and flexible Hon-Do pen. In this section, we explain how we simulate the characteristics of Hon-Do pen.
Hon-Do pen has the characteristics of stiffness and taper. Furthermore, the alum-Hsuan paper has the feature that water absorption is low and the ink does not diffuse easily. So, in sketching of Chinese Fine-Brushwork Painting, Hon-Do pen can generate clear and obvious lines. Lines seldom have the situation of diffusing. In the respect of drawing（運鋒）, artists often sketch with Chung Feng skill（中鋒）. It means that artists place the tip of pen in the middle without any slanting to make sound, strong, neat and smooth script. Therefore we use a single circle to simulate the contact region of Hon-Do pen on alum-Hsuan paper. We do not consider the situation of ink diffusion. We just take the stroke’s variation of width and color into consideration.
Figure 4.2 The footprints of circles (a) Thin (b) Tight (a)
(b)
Figure 4.3 The brush model
We define a circle which changes the radius along the variation of tangent line to represent the contact surface of Hon-Do pen on alum-Hsuan paper. This circle will move along a curve and sweep the footprints as the ink calligraphy on the canvas.
Consider Figure 4.2, when the contact circles are close enough, the whole footprints seems like a brush painting stroke. Figure 4.3 shows a curve depicted by Bezier polynomial. We may define a contact circle C with radius r, P is the point generated i by substituting u into the Bezier polynomial i Q(u) , S1 and S2 are the thresholds of the curve’s two extremities to restrict the circle C’s variation on them.
We define:
2. base: the minimum radius of circle C. It presents the minimum width of the Hon-Do pen stroke.
3. start: the initial radius of circle C. It presents the initial pressure when drawing.
4. θ : the included angle of adjacent points’ tangent lines as shown in Figure 4.4.
5. k: an adjustable degree for angle θ .
6. m: the u’s unit variation. It means the disparity between u and i u . i+1
Figure 4.4 The ө representation
Consider Eq.4.1, when the value of u lies between i S1 and S2, the width of line depends on the variation of the line’s curvature. In real painting, artists usually draw heavier to generate thicker stroke when the line’s curvature becomes larger. So, we produce this effect by using the variable θ . The larger the included angle θ the greater the line’s curvature. So we will obtain wider lines, and vice versa. We can also adjust the influence of angle θ by the parameter k. Then, we let the equation divide by m to decrease the effect of different disparity between u and i u . If we assign i+1 smaller disparity of u, it will cause smaller θ . That will affect radius r indirectly. In other words, if parameter m does not exist, we may get different widths of stroke when the u’s unit variation is different. But it is not a reasonable result, we use the variable m to avoid it. When u is smaller than S1, we simulate the effect of initial
S
between P and 0 P are generated by blending start and S1 r . On the other hand, S1 when u is greater than S2, we decrease radius r from r to 0. It simulates the S2 ending of the stroke when painting.
So, after all parameters are determined, we set a very small disparity of u. Then we can draw many circles with various radii and let them as the footprints of the ink calligraphy to obtain the goal of simulating Hon-Do pen style.
In respect of color’s variation, artists always use brushes full of ink to sketch. So we will not obtain the effect of dried pen（乾筆）. But the ink still could distribute non-uniformly, and it will fade away gradually. In our proposed model, we obtain this effect by adjusting the intensity of the color. The user can choose the color they like.
Then, the system will transform it from the RGB domain to HSI domain. When drawing, we raise the intensity of the color and let it change pale smoothly. In order to make non-uniform color, we cut circle C into several parts, then apply colors with slightly different intensity to each part. Then, we use a smooth filter on the stroke to make it vary smoothly and naturally. Figure 4.5 shows results of our proposed brush model.
Figure 4.5 Examples of brush model (a) Original line
(b) Stroke by start=1.0, base=2.0, k=0.2, S1=0.2, S2=0.8
(c) Stroke by start=8.0, base=3.0, k=0.4, S1=0.1, S2=0.8
4.2 Sketching the Contour on Petal
In this section, we introduce the procedure for sketching the contours on flower’s petals according to the input image.
4.2.1 Edge Extraction
We use Canny edge detector to find the silhouettes in an image. The Canny edge detector is a well-known algorithm. It uses the gradient of color in an image to determine the location of lines and find continuous and single-pixel-wide lines. User may input three parameters to decide the output line’s delicacy and importance, as shown in Table 4.1. The full and detail algorithm can be found in [6].
Table 4.1 Parameters in Canny Edge Detector
In our proposed system, the user can use proper parametric values to exclude unnecessary detail information and obtain appropriate lines. Since the goal of this step is to sketch the contours, the lines which can show the outlines of petals have higher priority.
Parameter Function
δ The parameter of Gaussian smooth filter used to eliminate the noise in input image.
high The upper threshold. If the magnitude of a pixel is larger than high, it will be considered as an edge pixel.
low The lower threshold. If the magnitude of a pixel is smaller than low, it will be considered as a non-edge pixel. If the value of a pixel is between high and low, it is determined by the connection.
(a) Original image (b)δ=1.2,low=0.6,high=0.9 (c)δ=1.2,low=0.6,high=0.8 Figure 4.6 Examples of Canny Edge Detector
4.2.2 Stroke Definition
After the edge extraction, we obtain a group of edge pixels. But we can not decide which points will form a stroke. In Chinese Fine-Brushwork Painting, artist sketches a line once based on the shape of the object. Consider Figure 4.7, an artist will draw a stroke for a petal, or draw two symmetric strokes for it, etc. Therefore, we have to group these pixels into strokes.
(a) One stroke for a petal (b) Two strokes for a petal Figure 4.7 Examples of strokes.
We first make judgments of connection to link points in neighborhood. We hypothesize that the connected pixels belong to one stroke. To avoid the disconnections generated by Canny Edge detector, we assume pixel in a 5×5 window to be connected. We choose a 5×5 window to define connection by our experience.
The window of 3×3 squares makes too many disconnected strokes, and the 7×7 window causes some unnecessary connections. But, the connection defined by a 5×5 window still causes many unreasonable linking lines as shown in Figure 4.8(a). Pixels in the same color are considered as a connection. So we have to cut off or eliminate some unsuitable connections according to the following rules:
1. The pixel whose links are less than X pixels will be eliminated. The default value of X is 30, which can be changed by user. Its goal is to avoid too short lines or lines interfered by noise.
2. The lines which are closed connected into a stroke in pistil area will be eliminated.
Pistil area is a region selected by user. Because the procedure of pistils is different from that of petals, we eliminate the pistil’s lines to avoid them being redrawn.
The method of the pistils will be discussed in section 4.4.
The method of the pistils will be discussed in section 4.4.