Google’s new open-source algorithm for image compression manages to reduce the size of a JPEG file by up to 35 percent without compromising image quality, and is compatible with all browsers and image editing programs.
This new JPEG encoder is called Guetzli, and its magic resides in a step of the compression process called “quantization”, which is when the algorithm tries to balance the reduction of size and loss of image quality.
Guetzli uses what Google calls a “psycho-visual model” to reduce file size by taking into account color perception and “visual masking” much more advanced than other algorithms. However, this has a small cost and is that Guetzli is slower than other compression methods.
There are numerous ways to adjust JPEG image quality and file size, but the so-called Guetzli focuses on the quantification stage of compression. For those who do not know, quantification is a process that attempts to reduce a large amount of disordered data (ie data difficult to compress) into ordered data. This process, at the same time, often reduces color failures.
Here we can see some examples of how Guetzli works:
A cutout of a cable hanging in front of a blue sky. From left to right: original image, standard compression libjpeg and compression Guetzli.
A cut out of a cat’s eye. From left to right: original image, standard compression libjpeg and compression Guetzli
As we can see, the new system works much better than the standard libjpeg system, although it takes longer to produce the result. While Guetzli’s main use will be to reduce JPEG file size, Google Research recognizes that it will also use it to increase its quality. When comparing the images coded with Guetzli and libjpeg (a most popular open-source encoder), 75 percent of the images look better with the first.
Guetzli is open source, which is good news as it could help reduce the size of images posted on the Internet, which in the end implies lower loading times and less data consumption.