Since there are multiple techniques for supersampling, I was wondering if there is any one technique that is objectively better than the others? Also, would the image content influence the effectiveness of the supersampling based on different techniques?

@CharlesLiu02 Good questions without short answers!

Sampling, filtering and resampling are theoretically and practically deep topics, and there are many approaches with different tradeoffs. You can take entire courses on these topics. In this course, though, you are just responsible for the core lecture part, so these slides at the end are optional extension.

That said, a few comments for you and other interested students:

• NxN supersampling, the way we ask you to implement it in the first assignment, has a lot to recommend it, including simplicity and efficiency.
• Randomizing the locations of the samples, as depicted in the previous slide, is often very helpful, because it can eliminate aliasing completely (why?), though with the the tradeoff of adding visual noise. We will study this later in the course when we talk about Monte Carlo Integration, and apply it in photorealistic image synthesis (assignment 3-2)
• As shown on the next slide, weighting the supersamples during averaging down can have a big effect. A theoretically good weighting function can be a windowed 2D sinc. But even cubic polynomial weighting functions can work well with the right coefficients.
• A pointer to great depth on some of these topics is the book on Fourier Transforms by Osgood that is linked on the reading page.

Come chat in office hours if this topic interests you!