Here's the issue I think gweinberg has: Imagine you encode an image to JPEG, with the "quality slider" cranked all the way to 11 (or whatever the maximum is). N

Here's the issue I think gweinberg has: Imagine you encode an image to JPEG, with the "quality slider" cranked all the way to 11 (or whatever the maximum is). Now, the DCT blocks are tiny and the compression is super inefficient. But it exactly reproduces the input image, and you may say it's lossless. The same image encoded with PNG (deflate) might be way smaller, but that still does not change the fact that JPEG losslessly encodes this image.