Actually it's not that new of an idea -- it goes back to at least 1996. I strongly suggest reading Microsoft Research's Foveated 3D Graphics. (click 'View publication' and read the full pdf)
In a nutshell, human visual acuity is different near the center of vision compared to the peripheral. The middle of the eye is packed more tightly with cone receptors making it more sensitive to detail and color. In contrast, humans are quite sensitive to motion in their peripheral vision. Evolutionarily speaking, it's a good idea to maybe take a look at that tiger sneaking up on you from the side. Your motion perception will trigger a shift in your gaze so you can make sense of exactly what's trying to eat you.
The most obvious thing that's different is your ability to perceive detail (angular size). Numerous things can be adjusted with the rendering to economize on detail.
- Resolution can be reduced.
- Texture sizes can be reduced.
- Geometric (polygonal) detail can be reduced
Color perception is reduced in the periphery as well, therefore:
- Texture maps with lower bit-depth can be used.
- Precision of color calculations can be reduced.
CAUTION: There are nuances to making effective use of all of these techniques. You have to be especially careful not to change the silhouette of the objects because the outer visual region is actually more sensitive to the motion differences that may be perceived by an edge jumping around. It may also be tempting to turn off antialiasing, but artifacts introduced by crawling pixels are tantamount to motion and draw the eye to these artifacts. Anti-aliasing sometimes needs to be increased negating the effects of foveated rendering. A blended blurring technique is used that allows seamless transition between regions rendered at high-resolution and low-resolution. It's a complex subject -- read the paper.
