An image compression test

By Charlie Loyd, 2006-04-28.

From time to time I say something about wanting better image compression for a photo and someone suggests PNG. I want to show why PNG, which is great for some kinds of images, is usually a bad value for sampled (i.e., photographed, scanned, 3D rendered, etc.) pictures.

There are three PNGs below. They come from the same hard-to-compress crop of a photo I took today. One of them is a direct conversion from a 111,781-byte JPEG. Another is a conversion from a 102,808-byte JPEG 2000. Another has only ever been a lossless file; most lately a 817,663-byte PNG. (I compressed the files with Graphic Converter 5.3.7X, which in retrospect wasn’t so hot at a certain one of the codecs.)

For regular posting, when not just making a point, I’d be more likely to save a 525×720 image as a 200 or 300, not 100, kilobyte file, so even if you spot the difference right away, I’m not sure you’d usually be able to.

This is just a quick draft of a real perceptual quality test, obviously. Please pass along suggestions for a fairer and more comprehensive version.