The root cause of aliasing is that continuous geometry cannot be correctly expressed with discrete pixels, resulting in unsmooth edges.
However, the more pixels and the higher the resolution, the less obvious the aliasing phenomenon will be. So the first way to anti-aliasing is to use a higher resolution during rendering calculations. This method is called Super Sample Anti-aliasing (SSAA), but this method obviously has a lot of performance overhead.
As shown in the figure on the left, it is usually judged whether a point belongs to a certain triangular patch, based on whether the center point of the pixel is covered, and this center point is also called a sampling point. Increase the number of sampling points, and average the color according to the number of covered sampling points, as shown in the figure on the right, this is the MSAA method.
MSAA uses continuous geometric information, and GBuffer does not have continuous geometric information in deferred rendering, so MSAA is usually not suitable for deferred rendering
FXAA uses image processing algorithms to achieve anti-aliasing effects by detecting and smoothing the edges of the image.
The principle of TAA is roughly the same as MSAA, and each pixel has multiple sampling points. But different from MSAA, TAA (Temporal Anti-Aliasing) integrates the data of historical frames to achieve anti-aliasing, which will spread multiple samples of each pixel to multiple frames, and the relative overhead is much smaller.