What are some antialiasing technologies used in consumer technologies i.e. Photoshop vs. high-resolution games? I've read that supersampling is fairly high-cost (as you might guess) and multisampling is a lower-cost version that uses edge detection to figure out which pixels to blur, but where might it be better to use one or the other (or one of the MSAA variations that exist)?
tpan496
I remember seeing things like 4x/8x/16x anti-aliasing optinos in 3D games. Is this the same algorithm used in those games to deal with anti-aliasing?
keenan
@merc The techniques used in rasterization (i.e., taking a list of abstract primitives and turning them into pixels) and image processing (i.e., taking an existing image and upsampling/downsampling it) are related but different. For rasterization, antialiasing is typically achieved using the methods we discuss in these slides, such as supersampling. Image processing algorithms will use various interpolation schemes (bilinear, bicubic, etc.) in order to 'guess' what more samples of an existing image would look like. One could in principle take a low-resolution rasterized image and upsample it in order to get a higher-resolution one, but this wouldn't help with aliasing, which is a result of under-sampling the original signal. For instance, if there were triangles that were missed by the original sample points, they still won't be seen in the upsampled image.
keenan
@tpan496 Yep, that's basically it. Basically those games are just telling the hardware to use a given supersampling pattern, and then downsample to screen resolution. (Ok, in real hardware it's a bit more complicated then this, but that's the gist. :-))
What are some antialiasing technologies used in consumer technologies i.e. Photoshop vs. high-resolution games? I've read that supersampling is fairly high-cost (as you might guess) and multisampling is a lower-cost version that uses edge detection to figure out which pixels to blur, but where might it be better to use one or the other (or one of the MSAA variations that exist)?
I remember seeing things like 4x/8x/16x anti-aliasing optinos in 3D games. Is this the same algorithm used in those games to deal with anti-aliasing?
@merc The techniques used in rasterization (i.e., taking a list of abstract primitives and turning them into pixels) and image processing (i.e., taking an existing image and upsampling/downsampling it) are related but different. For rasterization, antialiasing is typically achieved using the methods we discuss in these slides, such as supersampling. Image processing algorithms will use various interpolation schemes (bilinear, bicubic, etc.) in order to 'guess' what more samples of an existing image would look like. One could in principle take a low-resolution rasterized image and upsample it in order to get a higher-resolution one, but this wouldn't help with aliasing, which is a result of under-sampling the original signal. For instance, if there were triangles that were missed by the original sample points, they still won't be seen in the upsampled image.
@tpan496 Yep, that's basically it. Basically those games are just telling the hardware to use a given supersampling pattern, and then downsample to screen resolution. (Ok, in real hardware it's a bit more complicated then this, but that's the gist. :-))