Perceptual Encoding

[BezPowell]

This is part feature request and part query as to whether the requested feature would be usefully incorporated into zune-image itself, or better implemented in an external crate.

Perceptual image encoding is a technique for trying to minimize file size, while maintaining a given perceptual quality. It utilises multiple encodes and SSIM comparison to select the smallest quality setting for a lossy codec that falls within an acceptable perceptual deviation from the source image. This is implemented by such libraries as JPEG Archive and pio and is useful as the same codec quality setting does not produce the same perceptual deviation across different images (e.g. blurry vs high detail) and, therefore, selecting a single quality setting for all images in a corpus won't guarantee an optimal quality vs file size ratio for each individual image.

While it is possible to implement this yourself, a unified interface that accepts different input and output formats would be extremely useful as part of the zune library, or as an external crate that interfaces with it.

My current thought is to have a crack at writing a crate myself that perhaps uses zune's Image trait etc, but I thought I would post something on here first in case there's interest in incorporating this directly into zune, or someone else is already working on a similar project.