Forum_account wrote:

After rotating the points we see that the center of the new pixel lies between four points from the original image. We can do a bi-linear interpolation to get the color value of the new pixel.

Not only is this easier to compute, it's more correct. The first method assumes that every point has the same color as the nearest sampled point, this produces the stair-step interpolation function and the idea of "square pixels". This is a bad method of interpolation because it's probably incorrect. Not only that, but we then subject the image to another interpolation when we take the weighted average based on area.

Oh man. How did this guy become BYOND god back in the day? It's like, everything I read that he wrote starts out with really solid theory, but then somehow jumps to the polar opposite conclusion that would be indicated by logic.