Google has developed an open-source JPEG encoder capable of reducing size of JPEG files by up to 35 per cent.
Google’s new JPEGs are highly compatible with all exiting browsers, devices, photo editing apps and the new JPEG standard. The new JPEG encoder is named Guetzli and is freely available to download from Github.
There are countless ways to modify JPEG image quality and file size, but Guetzli pivots around quantization stage of compression. Quantization is a process to reduce large amount of unkempt data that is difficult to compress into a sequential arrangement that can be compressed easily. In JPEG encoding, quantization gently reduces colour gradients to single blocks of colour and highlights the smaller details to the maximum possible enhancement. By maintaining the file size to a constant, image quality can be significantly improved.
According to Google Research, Guetzli uses a new psychovisual model. ‘Psychovisual’ refers to human vision based processing system. In simpler terms, while considering smaller details in an image, colour perception is approximated and visual masking is achieved with assistance of a thorough and detailed observation to emboss them.
Google claims that the visual quality of JPEG images is directly correlated to its multi-stage compression process: colour space transform, discrete cosint transformation and quantization. Guetzli strikes a balance between minimal loss and file size by employing a search algorithm that tries to overcome the difference between the psychovisual models. However, Guetzli creates smaller image file sizes; the trade-off is that these search algorithms take significantly longer to create compressed images than currently available methods.
Google Research reckons that Guetzli can also be utilized to increase the perceived quality of JPEGs while keeping the file size the same. Comparing Guetzli–encoded images to other popular open source encoder – the odds are in favour of Guetzli by 75 per cent. This implies that Guetzli is reasonably close to human perception at high quality levels.