This was brought up here:

https://docs.google.com/document/d/1_lieRiHBBj5khV7kpMqjfrPfNSPUAS2HfTmRs2VVQuQ/edit?ts=5a90da63#heading=h.nxwsi5nay3xv

There is potentially a lot of meta data in compressed images that is not used by Chromium. They include comments and EXIF data. We'd like to determine how much data is "wasted" in the sense that we didn't use it to affect rendering.

One way of doing this is to add code to all the subclasses of ImageDecoder to detect the meta data we ignore, since it varies depending on the image format. Some things are simple to detect, like comments, but there are also some things that are a little more complicated, if we want to dig deeper. For example, a PNG may have conflicting color profile information, so some of that is ignored. And IIRC, there are many different color profiles built with tables that are intended to represent SRGB. If we treat those as the same (I don't remember for sure how we handle that), we could send the smallest version of it.