Terrain Synthesis from Digital Elevation Models
Zhou, H.; Sun, J.; Turk, G.; Rehg, J.M.;Visualization and Computer Graphics, IEEE Transactions on
Volume 13, Issue 4, July-Aug. 2007 Page(s):834 - 848
Digital Object Identifier 10.1109/TVCG.2007.1027
(available through IEEE Xplore)
I thought the article was well-motivated. Anyone who has tried to use a 3D package to generate terrain can relate to the authors' starting point. Personally, I have played around with Bryce and while it does creatively provide a painting metaphor for adding user-generated detail to procedurally generated terrain, it is ultimately very difficult to "paint" realistic-looking terrain.
ReplyDeleteThe article was interesting and well-founded. However, the authors' objective, "visually compelling terrain," was neither well-defined nor easily quantifiable. Moreover, their conclusion does not include a discussion of whether or not visually compelling terrain was acheived. The image results they included were impressive (particularly the lambda-shaped Grand Canyon), but it seemed to me that their visual appeal relied heavily on Terragen's procedural texture generation, which probably suffers from the same core issues as traditional procedural terrain synthesis (that is, lack of user control over local features in spite of robust control over global parameters). It seems like an interesting avenue of future work would involve a system of example-based texture synthesis to complement the example-based terrain synthesis that the authors explored.
Lastly, the authors make an interesting fundamental assumption that "important" aspects of the user sketch map can be coded in topological, so-called "curvilinear" components (i.e. branches, endpoints and the stretches that connect them.) However it seems that there are important terrain features that cannot be encoded by such curvilinear entities. Off the top of my head, large expanses of things would be hard to represent. This is evident in the authors' Middle Earth example: the large bodies of water had to be masked off manually because the pattern identification algorithm couldn't handle them. Perhaps future work could involve a more robust algorithm to automate more of the process.
Overall I thought this was a very good paper. I've never worked with any of these kind of terrain generation tools, so I don't have any personal experiences in the difficulty of producing realistic terrain. However, the paper does do a good job of showing the shortcomings of previous methods of procedurally generating this sort of terrain. The methods they used, and others that they drew upon, were a bit difficult to follow as someone with a limited knowledge of this particular field, but that's not unexpected.
ReplyDeleteThe examples given show some very interesting results in applying different elevation models to the specified ridgelines. It does seem to do a very good job of maintaining the subtle features. I'm referring to the physical characteristics of the terrain, since the textures are not part of this. For example, the authors bring attention to the curtain-like folds of the mountain sides in first half-life mountain range. A tool like this could be very useful in generating dramatically detailed terrain where the specific features are important (such as in a game with an established geography. The Middle Earth example was particularly good for demonstrating this capability).
One limitation of this system is that it is based solely on ridgelines with no regard to relative elevation or other features that might be contained in the elevation models. The system currently fills in space beyond the generated ridges with nonfeature patches, which are effective when the ridgeline is the only important feature. However, it is not difficult to imagine a setting where the creator may want a similar level of control of less pronounced features present in the same elevation models. This system also seems to be incapable of generating any sort of plateau. The authors do make mention of future work in adding constraints to incorporate desired elevations at specific points, which may make some of this possible.
Considering the fact that this is my first time ever dealing with graphics, the
ReplyDeleteidea of being able to render more realistic landscape models while
adding your own twist to it fascinates me. Anytime someone or something
takes a step towards making something seem more realistic, the graphics field
is expanding. The idea of being able to merge realistic qualities of a landscape
with user-generated qualities is novel in that it gives the user more control.
Although I found this information interesting, I did find it difficult to understand. If
the author would have had more illustrations or examples, I would
have been able to grasp the information better. Once again, I am new to
computer graphics, so I am not familiar with some of the techniques or
terms mentioned in the article. At the same time, however, the format of the paper
was helpful. The way the author outlined his points helped me follow him better.
Since I don't fully understand the methods presented by the author, I can not suggest
improvements to those methods. I can say, however, that better terrain modeling
will help improve the authenticity of video/computer games and movies where computer
graphics are used. As people began to find ways to make objects look more
realistic, the marketability of ideas, such as games and movies, where graphics are
essential increases which increase revenue.
I was very impressed by the quality of terrain that was generated using this new method. As a long time gamer, I am tired of seeing uniform looking terrain because the same texture was repeated over and over again. It will be nice to see more realistic looking terrain in the future if this method or something similar is adopted.
ReplyDeleteI found the detailed explanations of the mathematical models to be way over my head, so I kind of skimmed over them. But the paper was obviously intended for people with Computer Graphics backgrounds, so the paper was well written in that regard.
The authors themselves admit that their model fails on terrain with too little or too many curvilinear features. Desert regions are an example where they would get poor results.
All sorts of applications would benefit from this, including games and software like Bryce. However, since the model doesn't address if the model is geologically accurate, I am not sure if it could be used for flight simulators or any other real world modelling.
I found the paper really interesting since it covers a new technique for a field that currently has a lot of relevant applications. I am not too familiar with terrain synthesis software, but I am aware that it is still in the developmental stages. One thing that is really interesting is their approach to implement this new method since they divide the sources of information into separate branches where they consider both the user sketch of the terrain and real terrain data (height map) in order to obtain more realistic results.
ReplyDeleteIn general the paper was well written. The level of information is advanced; however, they give a very detailed explanation of the algorithms used and how the stages of their methods are structured, this makes the paper easier to understand. It seems that they have made a lot of effort in terms of efficiency, since they treat the features of the terrain as nodes and they consider the paths of the terrain as edges in order to traverse the terrain information in a breadth-first search fashion (acyclic graph). They also allow copying patches from the existing terrain and matching pixels that need to be filled. However, the algorithm is based on the polygon-breaking algorithm developed by Chang in order to get the terrain features, they argue that several improvements have been made by since there is not detailed explanation on this algorithm it might be a good place to start when looking for improvements.
Finally, I consider that this field of research has a lot of applications. Computer graphics is currently being used in simulators, training, movies and computer games. But this technique can also have a significant application for topographic studies and engineering.
- Romulo Manzano
I find the concept interesting. Being able to draw a simple shape and have the software try and make it realistic is a much easier way to go about terrian generation. I don't have any personal experience in terrian generation other than a map editor for a game called savage which used a gray scale image with darkness related to elevation to generate the terrian, and I thought that worked well enough.
ReplyDeleteThe majoity of this paper was difficult to understand as I am not familiar with the terms being used. The writers did a poor job at explaining terms, but it is a peer review type of journal so that is to be expected.
This will be great to use in creating terrian that does not exist, but you want to have the realistic features of some known terrian. Games and movies seem like the most applicable area.
a) what did you find interesting or novel about the paper?
ReplyDeleteThe most interesting thing about the article was their aim of making the method easy to use for the end-user. In my experience, research papers are lacking in that area.
b) what aspects of the paper were most difficult to understand?
Fish elevation? Not being versed in most of the foundations of the article, it was hard to understand the . The article clearly can't explain every aspect of their research in minute detail. They have to rely on the reader's knowledge, so I don't expect to be able to understand the entire article (yet!).
c) was the paper well written?
Sure, no qualms. Third-person passive, clear explanations based on assumption on reader's experience. It's at least as well-written as our textbook.
d) could the methods have been improved?
Of course. No method is perfect. As with any GIGO-based method (of example-based if you prefer), the quality of the result depends on the quality of the input. I don't understand the maths behind this method well enough to offer any suggestions for improvements.
In high school I did a project that had the same goal as this paper, although extremely trivial compared to this article's approach. I used a greyscale topographical map of part of Arizona as a displacement mesh on a flat plane, and then manually textured it. My method can definitely be improved!
e) what possible applications does this have?
I don't think this has many/any applications in science and engineering. The main benefit of this method is the ability to VERY rapidly produce terrains. Once the samples are "learned", new terrains can be generated with simple 2D sketches.
Artists in any area could use this method to quickly generate terrains for their work, or at least for inspiration.