Right, but a his poorly constructed example might be used against his general argument that there is no way to recover the content he purchased with his hard earned money. Of course, his example of being able recover the content itself is a weak example because not everyone will have the skillset to get around DRM.
So you start with a 4MB music file with DRM and lossy compression. After stripping the DRM, you've now got a 40MB music file at the same quality. If you want to get back down to 4MB, you must lose quality.
It wouldn't be a problem for jpeg (though I believe it is for mp3?). In jpeg the quantisation step controls the degradation in quality - you can run it through a 2nd time with the same quantisation without losing further quality.
Indeed, since I posted that comment I've been looking into it further. I read on wikipedia [0] that:
The DCT based form cannot guarantee that encoder input would exactly match decoder output since the Inverse DCT is not rigorously defined
I don't get that - I thought the inverse of the DCT was just the linear combination of the components.
So...I've been playing around with it myself to see what happens (using graphicsmagick).
I converted an image to jpg (quality 70) and then bmp -> jpg again. There are definitely some minor artefacts.
I then iterated that step jpg -> bmp -> jpg (quality 70) a dozen times. These further iterations don't add any additional artefacts. So it sort of gets into a stable state.
EDIT: reading that quote from wikipedia again it looks as though it's just saying that jpeg is lossless (ie, we apply quantisation).
These guys used to let you run your own images through their filter. It was great for spotting even really good shoops. http://www.errorlevelanalysis.com/
How?