Google is no stranger to the world of image re-encoding, and this time around, they’ve created an open-source encoder called Guetzli that’s billed as an alternative to JPEG. Guetzli is not only open-source and seamlessly usable in place of JPEG, but it tends to produce images that have a smaller file size and exhibit less of the artifacts that the JPEG format is notorious for, all while keeping the original resolution and almost all of the original detail of the image at hand. Guetzli uses complex algorithms applied over the JPEG process to better preserve the original image quality, but takes longer to create compressed images.
JPEG images are encoded in three stages; color space transform, discrete cosine transform, and quantization. The quantization part of the process, wherein an image’s building blocks are consolidated to a smaller number allowing for greater compression, is what Guetzli specifically targets. While JPEG is not shy about using extensive quantization to compress an image, Guetzli uses straight quantization much more sparingly, and instead uses smart algorithms to analyze the image and guess at which building blocks, like certain shades, hues, and visual details, are okay to ditch and will still leave the best approximation of the original image. This means that Guetzli’s version of the quanitzation process can remove just as many building blocks if not more, while retaining higher image fidelity by blending compatible building blocks and only making cuts where its algorithms tell it that there won’t be much impact on the image quality.
Google’s goal with Guetzli is not just to present a handy new way to handle image compression, of course; as with almost everything Google does, Guetzli is meant to be built upon and repurposed in a continuous fashion. The blog post announcing Guetzli says as much, billing it as a pioneer of psychovisual research in image processing; that is, helping machines to figure out what parts of an image could be dropped without the human eye perceiving big changes. Google hopes that Guetzli will spark further research in this area, but also that Guetzli will find usefulness in its current form by helping to make the internet a smoother-loading, better-looking place for all who use Guetzli or view images that it has been used on.
