[bob1029]

> Codecs like JPEG use a variable amount of quantization

The reason quantization works so well in JPEG is because of the DCT step and its energy compaction properties. This gets most of the coefficients near zero. I think without this transform you would be introducing a lot more noise in the final result.

At some point, we are going to end up re-implementing a thing approximating jpeg here. Colorspace convert, subsampling & DCT+quantization is most of the magic sauce.

[klodolph]

The singular value decomposition is doing the same thing as the DCT, sort of. It’s transforming the image into coefficients where the important coefficients are packed in one location. It’s worth applying similar techniques. You can see the demo where it’s truncating the coefficients.

Half the image codecs out there look like JPEG if you squint hard enough. That’s ok, “reimplementing a thing approximating” another codec is how a lot of marginal improvements have been made in the past, and if you add up enough marginal improvements, you end up with a competitive new codec in its own right.

The DCT isn’t magic, it just happens to work really well and requires little computational power. There’s a lot of possibilities for replacing the DCT with something else, especially now that we have more computational power to throw around.