Welcome to an overview of our DepthGen technology
DepthGen™ sits at the heart of our 2D to 3D conversion process and here at Prime Focus World we work hard to create proprietary tools like DepthGen to ensure that our conversion results are the cleanest and best looking around.
It addresses a common issue in the production of stereo images through conversion - namely occasional discrepancies between what you see in the left eye and the right eye. If you’ve watched 3D movies and sometimes found certain things hard to look at, it’s probably because of discrepancies like this. This is caused by poor conversion technology.
Our DepthGen technology helps us to create beautifully clean stereo for our clients, and gives them full creative and mathematical control over the process.
So how does it work?
A stereo image is produced by creating two slightly different views of an object - a different angle for each eye. DepthGen allows us to carefully shift pixels in an image to create this difference in angles, whilst minimizing the discrepancies that can be caused by this process.
So let’s focus on the characteristics of this pixel shifting operation.
DepthGen is driven by image processing operations known in computer science as ‘transforms’ and ‘filters’. Any time you want to shift pixels around in an image, you’ll need these functions and a way to control them.
The functions are controlled by a ‘disparity map’. Disparity maps shouldn’t be confused with ‘Zdepths’, which look similar but are very different. Making edits to the disparity map dials up or down the strength of the functions.
So, if the new eye angle we’re trying to create for a foreground object has it pointing further to the right, objects in the background should parallax to the left. As the object moves, the function is pushing some pixels apart, and crunching others together.
Where the pixels are pushed apart, the function has to fill in the resulting gap with new pixels. How the function creates those new pixels is crucial. It’s very important to the cleanliness of the image that these newly created pixels look very similar in both eyes - although they can’t look exactly the same.
Developers who code these functions have several options including;
- simply repeating the pixel across the gap
- interpolating the pixel across the gap
- or performing a weighted bias to fill the gap
However, in stereo these tend to not look so good. Any difference between what’s visible in each eye will be uncomfortable to the viewer and should be minimised.
Some off the shelf functions are included in popular compositing packages, but they can exhibit what we call ‘warping discrepancies’. If we compare the warp error produced in each eye, we can tell that this will cause extreme visual discomfort in stereo.
Using commercially available functions is cheaper than developing and maintaining your own but this is a false economy. The errors can be cleaned up by hand, but this takes too much time and manpower to be economically viable.
And because this is such a common operation in conversion, these artefacts appear like a stereo noise all over the final converted result - if you’ve seen buzzing edges in some stereo movies, it’s most likely due to this type of discrepancy.
At Prime Focus World, the results from our proprietary DepthGen algorithm are much cleaner straight out of the box, maintaining detail in the image and matching the edges of objects with much more fidelity to the original.
DepthGen sits at the heart of our conversion process, and is a vital part of providing our clients with the cleanest, sharpest, most flexible conversion results in the industry.
Contact Prime Focus World today and find out how DepthGen can improve your stereo conversion results.