In fact, the supersampling is not the exact method to meet the requirement of Nyquist frequency. It is more like a trick to mitigate the aliasing.
From the image processing perspective, it contains two steps.(1) Sampling using a higher resolution (twofold, fourfold) (2) Using a pooling with stride = size to downsample the image.
It is more like we only 'soften' the rigid edge of the shape, and neglect some characters with higher frequency. If the higher resolution sampling still does not meet the Nyquist frequency, the jaggy still exists but is just blurred.
Rui-Wang-813
I am actually a little bit confused on this super sampling idea. Why is it that there are multiple values within one pixel? Is it that actually a pixel covers a range of area and inside this area the values (color) change?
In fact, the supersampling is not the exact method to meet the requirement of Nyquist frequency. It is more like a trick to mitigate the aliasing.
From the image processing perspective, it contains two steps.(1) Sampling using a higher resolution (twofold, fourfold) (2) Using a pooling with stride = size to downsample the image.
It is more like we only 'soften' the rigid edge of the shape, and neglect some characters with higher frequency. If the higher resolution sampling still does not meet the Nyquist frequency, the jaggy still exists but is just blurred.
I am actually a little bit confused on this super sampling idea. Why is it that there are multiple values within one pixel? Is it that actually a pixel covers a range of area and inside this area the values (color) change?