Calculating a blurred image still requires the original image, but now you also need to sample each pixel from *multiple* source pixels in the original image, which is just additional calculations on top.
Also, there are many different types of blur. Some look good, most look bad. Cheap to compute blurs look terrible. Cinematic bokeh-producing blurs are really expensive to compute. Creating a “cinematic” DoF look requires the latter, but also needs to take depth into account at every step.
In short, blur is not just reduced resolution, it’s like turning each pixel into a translucent circle and having to blend them all together.
Latest Answers