Final Progress Report: under water chroma-keying
Patrick Burns :: Devin Hunt :: KURT Margenau :: Anil Rohatgi
The Effect Proposal
There are two effects in play in the film. First, the visual effect of weightlessness in otherwise 'normal' environment. This creates a feeling of surreality through the juxtaposition two environments. Secondly, some CG elements will be added to accent the weightless effects.
This is similar to the effects done in the Switchfoot music video "Stars" as well as Ricky Martin's video "She Bangs".
Click on the picture below to go to our critique on how these videos were made.
 |  |
Pre-Production
Download the PDF of the rough screenplay
Since a good majority of our source footage required that we be underwater during filming, extensive care was taken into carefully planning each shot in order to mimize the time we needed to be underwater and filming. Originally we inteded to film these shots in the Georgia Tech Diving Pool; however, due to an unfortunate incident at the CRC, the pool was no longer available. Luckily, we managed to find a local dive shop that allowed us to use their facilities to accomodate our underwater shooting. Presented below are sketches and videos showing the various stages of planning and pre-production that went into the making of this film.
Download a video of the story boards designed to map out timing
http://www.cc.gatech.edu/~trigger/animatic1.avi
Download a video of the animatic used to represent the salt shaker effect
http://www.prism.gatech.edu/~gtg618b/DVFX/final/animatictest.mov
Aside from conceptual shot planning, specialized hardware needed to be obtained and constructed in order to make the underwater shoots possible. An airtight underwater enclosure was bought in order to house the camera and prevent it from damage while filming, and an underwater tripod system was designed and built to hold the camera steady while filming. Below are images of these riggings.
 |  |
| Camera Housing | Concept Camera Rig |
Of course since all these initial designs were made to accomodate DV cameras and DV footage, by the end of the production they were all revised to film in HD.
Post Production
After several rounds of filming took place, the final footage was run through a series of programs written in Matlab and the output videos were then composited in Adobe Premiere v.2.0. The initial pipeline from filming to completion can be seen at the link below.
http://www.cc.gatech.edu/~trigger/DVFX_Pipeline.pdf
Also, an initial report describing how the code module would work was created and can be seen at the following link.
http://www.cc.gatech.edu/~trigger/Code_Modular_Overview.pdf
However, after some serious testing and implimentation ( over 15 matlab scripts were written to test various algorithms), the segmentation and color correction algorithms were changed drastically from the initial proposal. The figure below demonstrates all of the stages used in creating the final production.
Color Correction:
First step, is color correction and image cleanup.
Step 1 is to remove the noise. this is done both by running each frame through a median and mean filter to remove high freq noise.
Next the color is altered by performing a histogram equalizaton over each of the RGB planes of the input image. This spreads the originaly tinted blue footage over the entire range of values, making the colors look rich and un-tinted.
Lastly manual values for offset in the R G and B plane were added, along with a variable value for darkness. this was done as a last stage tweaking.
Segmentation
The input to the segmentation algorithm is the newly color corrected movie.
Each frame of the movie is then processed, first by splitting it into its R G B planes
Each plane is run through a median and mean filter to remover high freq noise
The algorithm then compares each pixel in the G and B planes. Assuming the background was a primary color... say blue... the pixil values will have a large difference in each plane. Pixels with a great difference are colored black. Others are colored white. This makes the mask.
The mask is then output as a binary image. Then the vales from the red plane are amplified and added back into the mask to fill
in holes and preserve a low level of background noise from the source footage. This produces a grey-scale mask.
Finally an opening morphological operation is run on the greyscale mask to remove holes.
This video is output as the mask for a shot.
Compositing
The next step to post production was to take all of the CG elements and foreground objects and composite them together with the color corrected footage and background plates. The background plates were all filmed in California Pizza Kitchen located in Atlantic Station. If you look closely you will also notice that the people filling the background plates are us as well as shown in the figure below. Below are images of all the different layers and composite elements used to comprise the final footage. All of these layers were imported, synchronized, and composited using Adobe Premiere v.2
 |  |  |
| CG salt shaker | CG bowls and dollar | CG water (last shot) |
 |  |
| Background plate | Foreground table |
All of these elements combined made our awsome video!! You can watch the finished product at this link.
And also see the equally as awsome making of here.
http://www-static.cc.gatech.edu/dvfx/videos/dvfx2006.html
Link to this Page
- DVFX 2006 Groups last edited on 28 March 2006 at 1:15 pm by r52h135.res.gatech.edu