Anyone familiar with the past decade of work from Jim Cameron knows he’s had his share of problems with 3-D. His persistence, and his willingness to learn and improve, is what spoiled us with the quality 3-D experience of Avatar. But when it comes to 3-D, a good part of the industry is only where Jim Cameron was 10 years ago. This is as true for the camera capture of 3-D as with the conversion of 2-D images to 3-D. With around 30 3-D releases produced annually, there will be much learning going-on, at the expense of audience discomfort. Such dismal 3-D eyesores as Clash of the Titans and Cats and Dogs underscore the need to understand the do’s and don’ts of conversion.
When converting an image, basic depth cues must be observed. These common-sense concepts include the placement of occluded images behind the object in the foreground. Objects that are similarly sized in nature may not appear similarly sized in the frame – this provides a visual clue as to the object’s depth. In a similar fashion, the relative height of same-sized objects provides clues to its depth. Motion also provides depth cues. If the camera is in motion, the further objects move more slowly than the objects that are closer. An aerial perspective will provide visual depth cues, where objects off in the distance may be bluish and hazy. If these simple clues are not honored by the 3-D converted version of the image, the conversion appears wrong. The explicit 2-D depth clues will not match the 3-D depth clues, leading to a degree of visual discomfort.
We interpret depth through a few mechanisms. One of these is the convergence of our eyes. Our eyes have to cross slightly to retain focus on nearby objects. This crossing is interpreted by our brain as the object’s distance. Binocular disparity is another factor that causes us to interpret the depth of an object. Binocular disparity is perhaps best explained as differences observed in the left eye and right eye versions of the object.
With this basic knowledge, one can convert 2-D images to 3-D. A matte is created for an object. The object is moved in terms of position in the frame for one or both eyes, playing upon ocular convergence. Paint work is then applied to fill in those parts of the image previously hidden. To add depth to features of an object, binocular disparity comes into play. Some differences in the object must be introduced to give it shape. This may be simple to explain, but it requires a lot of work when converting a complex frame.
Time costs money, and so most conversion houses tout shortcuts to reduce the cost of conversion. Some outfits base their entire process on one or two shortcuts. While shortcuts may produce an acceptable result under certain conditions, they will not produce good results in all cases. The only process that works everywhere is that previously described: matte, move, paint. It is the improper application of shortcuts that led to most of the bad conversion seen in recent movies.
One such shortcut is called “rubber sheet.” An enhanced luminance map is generated for the image, which is then interpreted by a computer into a 3-D depth map. Imagine a face on a flat plane (2-D), and then pulling features of the face forward from the plane through a digital transform. The face is still attached to the original plane, but now has dimension. By placing two cameras on the digitally transformed version of this face, a stereoscopic 3-D picture of the image can be created.
When looking at the features of the face, it may appear that the shortcut worked. No matte work, and no paint work, was required, but a 3-D effect was created. However, the face is still attached to its background. The edges of the face actually roll-back into its background, creating a unnatural look, and giving it the name “rubber sheet.” In effect, the rubber sheet technique introduces binocular disparity, without directly introducing ocular convergence.
Another common shortcut is to eliminate, or automate, the paint work. The conversion house creates the matte, moves the objects as required by the depth cues, but forgoes the cost of detail painting. This produces fuzzy, shimmering, or otherwise odd-looking edges that are uncomfortable at worst, and distracting at best, to watch.
Other cost-cutting shortcuts include the cardboard cutout look, where the process of matte, move, and paint are applied, but binocular disparities are not introduced to provide shape. Still other shortcuts include poor or no handling of image transparencies, poor management of depth cues resulting in miniaturization, and quick and dirty mattes that result in a composited look. There are many ways to cut corners in 3-D conversion, and, unfortunately, you will likely experience them all. Let’s hope that some day soon, directors and production professionals will come to understand the nature of shortcuts and better appreciate how to manage them in their 2-D to 3-D conversions.