AP projects 2015
Posts tagged image processing
As promised, we would give the summarized results for the two scanning sessions that we had: with CT scanners and with Artec Spider scanner.
Part of the results of the latter unexpectedly disappeared during the post-processing. The scan which promised the most for us, Hermione handle detail, was among the missing files. We hoped to get better results of the floral ornament and combine it with the CT-scan body, only the chosen comparison model and Harry survived.
As seen, for the comparison we chose the finest model we had. The lice-comb teeth were approximately 0.5 mm diameter with even smaller gaps between them. Due to this, the scanning technique used by Artec Spider could never achieve a proper result: too much was not visible, even with the precision of 0,05 mm. In other words, the grid which the range finding device projected could not be interpreted in the gaps and the result was a block with a texture instead of a comb. Moreover, looking at he scan of Harry we can see a big inconvenience for us: only the outer surface and the sections at the breaks were captured. Moreover, the cracked surface texture was not captured, because we got the file only in a mesh file.
Therefore, it would be expected that this problem would not be so apparent in CT-scans. This technique captures the sections of the object, instead of making an interpretation of surface. Just then these sections are interpreted into 3D files. However, the precision of 0,3 mm proved to be insufficient for the artifact we chose:
As seen, the result was a more consistent file, which could actually be printed. Nevertheless, it was far from what we would call sufficient. Expecting this, we also made micro-CT scans of the object. The sneak-peaks of the object in the lab itself looked very promising. Yet our and publicly available computers could not handle the size of the data set (over 2000 sections!) and could only give results in the lowest resolution, leaving us with the following model:
As seen in the picture, the separate teeth are clearly visible, although the main body is missing. This can be easily solved if the used computer has 16GB RAM, since we could get a proper model in Avizo a few moments before it crashed due to memory insufficiency.
To conclude, only the micro-CT scanner offered the sufficient results for the compared artifact. The Artec Spider is very interesting if surface detailing in necessary or if textures/colors have to be captured. However, if not enough scans are made and combined, you will get an object lacking details, thus resulting in incredible amount of work hours in post-processing. Another solution for this would be to make CT-scans and combine them with the Artec Spider scans only for the details.
After multiple emails and a Skype talk we finally acquired a trial version of Mimics. According to their representative, the program is mainly focused at medical uses. Most importantly how do bones, implants react to friction and temperature changes.
This got us interested, since this was also rather important in our project: we were using CT scans to determine the break-line positions and in the end also fill up the missing shards with (possibly) other materials. In other words, it would be very interesting to see how different connections between materials would influence the durability of the object. I must add, that this is only a presumption after a talk with their representative and we might not be able to go so deep in the subject due to the time limitations. Yet this could be very interesting as a research subject for future students.
Having only a week of work left till the presentation, we decided to only check what were the possibilities of the translation (CT-scans to .stl) procedure and if the results could be better than from Avizo.
The interface seemed clear, but more limited to what was offered at the latter program. It seemed actually very similar to already mentioned Seg3D, which is also focused on medical use.
After comparing multiple objects we came to conclusion that this program does not offer better translation. The meshing is coarser and even though the stepping is less visible, so are the break lines.
In conclusion, this program might offer higher possibilities going deep into material interaction(3-matic research), but for simple .stl translations Avizo is still the best option.
P.s. For post processing use MeshLab (open source!): there you can both reduce the fineness of the mesh and smooth it.
As promised a day ago we would keep you posted about our adventures with image processing software. Even though we did not receive the trial version of Mimics, Avizo provided a very pleasing outcome and it also read .dcm files.
The program itself is very user friendly and incorporates visual programming with automatized properties, thus giving the user just enough freedom to not crash his computer or make the process incomprehensible. In this sense it is very similar to 3Dslicer, yet Avizo has more options and more finesse in the details.
The final result is not yet perfect due to rather clear “stepping” in the final 3D model. This, according to our “informer” from the Industrial design engineering faculty, could be solved with Geomagic. However, now we can clearly state that the plan A mentioned in this post is actually possible.
To keep you up to date we would like to introduce of pending and active processes of the project.
Currently we are focusing on two main subjects: production techniques (Jorinde and Irene) and image processing (Sander and Kotryna). For the first one we plan to publish a page later today which will be updated as we gather information. For the second, we have already posted an overview, but since we are still waiting for the trial versions of some of the programs a second article will follow.
As expected, we are lagging with the step “CT-scans to .stl”. The reason for that is the low quality of already produced .stl files and inability to open the .dcm files.
Our plan A is to finish with it today or tomorrow depending on the outcomes from the trial versions of the program. Then on Thursday make the individual models of object “Harry” with production technique ideas. We will pitch them to each other on Friday morning, hopefully resulting in 2-3 final ideas which will be sent for production in the same afternoon. Most important notes for the preparation of these pitch files are:
- Thinking about the ability to produce
- Retaining the historical footprint
- Trying to make smart joints between different materials, since the current holes have very rounded corners (also a part of 1.)
Plan B would come into action if we cannot get the .dcm files to work or the quality of the scans is very low. This would lead to post-processing of already prepared 3D files of object “Dobby” to make the surfaces smoother and the break lines sharper. That would mean that we would loose a lot of data: including the micro-scans, which would be used for the comparison of all the scanners that we used. Yet we would still follow up the plan A from Thursday by shifting it 4-5 hours.
In conclusion, even though we are slightly behind primary schedule, we have received valuable insights in what programs to use and what to expect of them. Moreover, even if we loose the .dcm files, we would still be able to produce what we strove for in the beginning: just the final model would be less spectacular.
Keep your fingers crossed for us and till Thursday!
Directly after receiving the scans on Tuesday we jumped into processing them into 3D models. To keep it clear we used “Hagrid”(obj. 5) as an example for all of the programs.
As noted in the previous post, this process has multiple steps and in order to gain the highest level of detail, a lot of tweaking is necessary.
The usual procedure goes as follows:
(0. Changing the .ima or .dcm files into program compatible format. Most of our scans were made in .dcm format which was not compatible with multiple programs: so far we tried RenameMaster, which did not work)
- Loading the .dcm or .ima files into a 3D processing program. These file formats actually contain only 2D information: the sections of the object. In other words, the 3D model is an interpretation of multiple sections and therefore steps between them might be visible, if the resolution is not high enough.
- Selecting threshold and filtering the right information. Depending on the program this step might be automatized. If not, it might be very heavy on your computer. Therefore, a device with a good graphics card and 16GB RAM is advised (it would work on 6 or 8GB RAM, but it goes slow and tends to crash often).
- Loading the 3D file into a volume renderer to get an editable mesh (.stl).
To begin with, we started with Seg3D. This program did not want to read .dcm files, thus we only worked with test files, which were in .ima format. The interface was clear, but to extract minuscule details it needed a lot of filtering and playing with histograms. That was extremely hard on our computers (6-8GB RAM, 2.0-2.03GHz) and took over an hour to get a decent file. Moreover, the final result is given in .nrrd format which later has to be translated to .stl with the help of ImageVis3D. The file looked rather detailed in Seg3D, but the final .stl was worthless.
Later on, we received a tutorial from an past student of our supervisor Maaike. It suggested using DeVide. Unlike the previous program this one works on the basis of visual programming. Thus all of the steps can be easily retraced. This program can directly export to .stl reducing the possibility of getting a very rigid mesh, like with Seg3D. Unfortunately, the program did not want to work on our computers.
After this failure we contacted one of the researchers in the faculty of Industrial Design Engineering. He adviced to try out the following programs:
- 3D slicer (open source)
- Avizo (paid, evaluation copy available after contacting the firm)
- Mimics (paid, evaluation copy available after contacting the firm)
The first of the list (3Dslicer) proved to be very user friendly (although it did not read the .dcm files). The information is collected automatically after choosing a preset and is quite precise. One can also select if to smooth the surface: both outputs are interesting in form, with the edgy one as an expressive interpretation of a kitschy object of the past. If used for the final product, more mesh post-processing is necessary
To be continued…