Hairy Brushes
This paper presents an interesting approach to the problem of duplicating the nature of strokes written with a brush. However, I could wish for more detail on the specific painting method which the author chooses to try and duplicate. For example, only a picture of a computer-generated Sumi-e painting is given; it would be very interesting to see how thie computer-generated version could compare with one manually painted. Wihtout that comparison, it is, to me, unclear how accurate or intuitive a model the ``hairy brushes'' provide. The modular representation seems logical, though there are several effects that are neglected. For instance, bristles can acquire ink from neighbors; what about excess ink deposited on the paper? The pressure is presumed to vary only as a function of distance along the stroke; I would be curious to see if the pressure actually varies, in addition, across the brush. Clearly, there is a lot of future work to be done to amplify and determine if these represntation and algorithms are useful. I would also be curious to see how interactive such a system could be with modern technology.
Computer-Generated Pen-and-Ink Illustration
I liked this paper. It clearly presented a bounded problem, developed guidelines, and described the method for computer-generated pen-and-ink architectural drawings. The motivations given for such a system were clear and logical. The author did assume considerable knowledge of graphics systems, such as what a ``graphics rendering pipeline'' for photorealistic imagery is. However, even with that lack, the description of the differences required for pen-and-ink were clear. Not being familiar with architectural drawings, I can't speak as to the resemblence of the computer-generated images to hand-drawn ones. My only real complaint about the paper is that htere is, again, no comparison of the computer-generated with the manually drawn.
S. Strassmann, "Hairy Brushes":
An interesting article, dealing with a topic I have not heard much about: using a computer for artistic rendering, rather than rendering photographic realism. I am still a little confused about Strassmann's use of the bristle-color quantity in his brush model; it seems somewhat redundant in light of the other factors on which grayscale color could be based, such as pressure and the amount of ink left on the brush. Bristle color might have value as an intermediary quantity, but I wonder if it could be eliminated or subsumed under another aspect of the brush. Apart from this question, however, Strassmann's painting model in general seems quite well thought out, and his ideas for further work (better user interfaces, higher dimensional brushes, etc.) sound promising.
G. Winkenbach and D. Salesin, "Computer Generated Pen-and-Ink Illustration":
Another paper dealing with artistic rendering, as opposed to photorealistic rendering. The fact that Winkenbach and Salesin can produce such nice graphic results using a platform no more sophisticated than a Macintosh bodes well for their technique; removal of the need for special viewing/processing hardware is a plus. The pen-and-ink medium itself is an interesting and useful choice of representation. However, the paper seemed oddly uninformative at first glance. Although knowing the basics of pen-and-ink illustration is obviously important to the work, the authors seem to have spent a little too much time describing it at the expense of describing their own efforts. Putting the bulk of their procedural methods in an appendix at the very end of the paper did not help this initial impression; the first time I read the paper, I nearly mistook that section as a continuation of their references!
Hairy Brushes
This article discusses a computer graphics model for representing paint brushes as a collection of bristles. There are many parameters that are considered by this model. They are abstracted in four main objects: the Brush, a one-simensional array of bristles which are objects by themselves; the Stroke a set of attributes that change as a function of time or distance which make the computer strokes a more realistic representation of the real properties of a stroke made by hand; the Dip which represents the distribution of ink on the brush; and the Paper which renders the ink that comes off the brush/bristles.
All the object properties interact to create a painting model which is able to control ink quantity and color, their evolution with respect to a specific stroke, pressure, texture mapping and some other effects caused by the already mentioned.
There are too many degrees of freedom to be modeled. Although this is a simplified model, it can be useful to create other categories of paint-like media. Fast rendering, compact storage and more artistic appearance are among other benefits this model offer.
In general the article describes a model that considers a good and well defined set of variables / elements that influence the way ink is printed on paper. I think that all the elements considered, specially the modeling of the physical properties of the materials that interact in the generation of the picture, make this approach a realistic model of painting. Although a specific kind of technique of artistic painting (sumi-e) is the basis of this approach, I think that it can easily be extended to a more complicated model that pays attention to new elements and physical properties.
Computer-Generated Pen-and-Ink Illustration
This work is based on the use of principles of pen-and-ink illustration as the basis to create and computer-based implementation to automatically render real non-photorealistic scenes. Architectural forms is the direction chosen by this paper although it may be applied to other type of illustration.
This work is intended to use only the scene geometry, texture assignments for each surface in the scene and some lines to create the fields that form what they call "indication" in order to compute a non-photorealistic but very natural scene restricted to the use of pen-and-ink like elements.
Some principles of the technique for illustrating in pen and ink are described in the paper. They focus this description mainly by analyzing the properties of strokes, tones, textures and outlines. I think that it is complete description for their purposes, very useful when considering an implementation. Other principles are added by this paper, specially those related with a computer implementation, which can be extended according to the chosen technique
The main problem to solve is that despite the extra considerations for an automatic implementation, Pen-and-ink illustrations is a limiting medium. It is necessary to use combinations of individual strokes to suggest color, shading, etc. So it is necessary to create an overall impression of the desired effect by mixing the properties already mentioned.
Some fundamental differences and similarities with photorealistic imagery techniques are discussed: the dual nature of strokes and the combination of 2D and 3D information. Based on this, some changes are made with respect to a basic rendering system. The paper discusses the main changes. A brief description of the process of rendering is explained.
Visible surfaces and shadow polygons are computed. These polygons are projected to a Normalized Device Coordinated space. Every visible surface is rendered. The texture related to every surface is used to generate the strokes that give the desired texture and tone to the surface. The outline strokes are drawn by computing all the outline edges needed for the illustration.
In general this work describes a good set of principles, elements and considerations that have to be made when generating an illustration. I think it covered a complete set of them. Due to this, the pictures generated present good quality, detail, and in general power of illustration. Good properties are achieved: problems with change of resolution can be avoided (much better than pixel-oriented techniques), economy of illustration, complicated computations can be simplified by simpler principles.
1) S. Strassmann. Hairy brushes. In Computer Graphics Proceedings, ACM SIGGRAPH, pages 225--232, 1986.
This paper models paint brushes that can model a particular style of Japanese art known as bokkotsu sumi-e. The model described in this paper uses four major representational units: (a) brush, (b) stroke, (c) dip, (d) paper. The author emphasizes the need for using such a modular system of representation. The argument used by the author is that this allows the possibility of modifying the algorithms used to simulate the various operations associated with a particular representational unit without affecting any other unit. Also, it now becomes possible to change the attributes of a particular unit, leaving all other units unchanged. This allows the user to experiment with various dips for a given stroke and perfect the final painting. This is also very useful for editing incorrect strokes.
Since no special input hardware, such as a stylus with an acoustic (or sonic) tablet, was used, the inputs to the algorithm were obtained by using a mouse to specify control points which are used to generate a cubic spline interpolation. Also, such parameters as pressure at a point on the curve have to be specified using numbers, and the correct values to be chosen to create the desired effects are not necessarily obvious to a user. All this makes the system highly difficult to operate, i.e., it is not user friendly. Further, the algorithm uses only shades of grey to produce "color", so various coloring effects become impossible.
Another major problem with the algorithm is that some common operations such as polygon-filling are computationally very expensive because each pixel needs to be drawn in chronological order. This makes it impossible for the algorithm to function in real-time. The author himself agrees that the algorithm is very slow though he is very vague regarding the details: it takes a "minute or two" to render a single stroke. However, he claims that a large proportion of these computations are parallel in nature, leading to the possibility of the algorithm being very fast on a parallel computer: an unconvincing argument. This is especially true because the author says that the main motivation for this work was to create reproducible brush strokes that will allow paintings to be animated. However, he admits that "each frame (in the animation generated) took about one minute to render". This is a totally unacceptable frame-rate for any practical application. For all these rea sons, I think that the model proposed in this paper is a toy model and it would be rather useless in a practical environment.
2) G. Winkenbach and D. Salesin. Computer-generated pen-and-ink illustration. In Computer Graphics Proceedings, ACM SIGGRAPH, p 91--100, 1994.
This paper describes a method for rendering pen-and-ink illustrations on a monitor screen or some other output device. The authors begin by discussing the advantages of pen-and-ink illustrations over traditional methods used in computer graphics for "photorealistic" rendering. Photorealistic rendering methods produce images that are true to real-life, but it is often more useful to highlight some aspects of the image and this forms the basis for the usefulness of pen-and-ink images. Other advantages of pen-and-ink illustrations also exist; for instance, these images provide economy of expression in representing tones and textures, they possess qualities of simplicity, crispness and directness that are difficult to capture in other media, they blend nicely with text, requiring a single color of ink on a sheet of paper, and this makes them ideal for printing purposes, and, finally, they add an artistic touch to the document.
After discussing some of the fundamental principles of illustrating in pen and ink, the authors go on to investigate some of the major differences between the techniques used in traditional graphics methods of rendering and those that are required to render pen-and-ink illustrations. Two of these fundamental procedural differences are: (a) while in the former, renderings of texture and tone are completely independent, the latter require that tone and texture both be rendered by the same strokes, and (b) in the former, the information used for rendering is entirely three-dimensional, with the final projection to two dimensions largely a matter of sampling the rendered shades, whereas for the latter, 2D aspects of the projection must be integrated with the 3D information for creating a proper rendering. Thus, the authors used a basic graphics pipeline algorithm with a few changes to incorporate the above differences. One major feature that the authors added to their algorithm was the concept of prioritized stroke textures. This basically allows the final image generated by the algorithm to be resolution dependent in the sense that at very low resolutions, the low priority strokes are omitted, i.e., a lower level of overall detail is seen in the image. This improves clarity of the final image by reducing the amount of "blackness" (due to an excess of strokes). However, at higher resolutions, all possible strokes (low as well as high priority) are displayed in the image, thus increasing the level of detail and reducing the "brightness" of the image. This technique improves printing efficiency.
A key issue in any computer graphics application is the time taken by the algorithm to perform the required computations. For an algorithm to be useful, it has to operate in real-time. The authors do not discuss this issue in their paper apart from mentioning that it took the algorithm 30 minutes to compute and print an image consisting of 1043 polygons when the rendering of the image was done at a resolution of 600 dots per inch. It would be useful to compare this number with similar numbers for other applications. The authors also mention another problem with their algorithm which is that since their system uses randomness profusely, issues in frame-to-frame coherence arise in applications involving animation. For instance, large features that are random, such as the selection of bricks that are shaded, should not vary from frame to frame. However, more subtle features may be allowed to vary. Apart from these shortcomings, the paper was very instructive and highly interesting.
Stan Sclaroff
Created: Jan 21, 1997
Last Modified: Jan 30, 1997