I see in some of the technical papers, good practice for downsampling is to first pass the image through a Gaussian Filter and then take sampling to avoid problems like aliasing and so on.
However, I also see a lot of people simply use the bicubic smoothing function with a scale size below 1 (e.g. 1/2, 1/4) and so on.
What are the differences in characteristics between using these two methods? Which one is more preferable in the specific case of simulating a low res image from a high-res image?