Google has introduced a new ops library, Jpegli, that compresses images 35% more efficiently while maintaining high quality.
For years, Google researchers have been working on improving data compression (Brotli), audio (Lyra V2), and image compression (WebP). Jpegli is their latest project, aimed at improving the compression ratio of legacy JPEG files on systems where modern compression, such as WebP, may not be available or desirable.
According to the announcement on the official Google blogAmong the advantages of Jpegli:
- Compatibility The encoder and decoder are fully API/ABI compatible with libjpeg-turbo and MozJPEG.
- High quality. When working with images, more precise and «psycho-visually effective» calculations are performed, resulting in clearer images.
- Speed. Although Jpegli’s image quality and compression efficiency are better, its encoding speed is on par with libjpeg-turbo and MozJPEG. Thanks to this, web developers can easily integrate the new library into their workflows without sacrificing speed or memory consumption.
- Ability to encode an image with 10 or more bits per color component. Traditional libraries encode images using 8 bits per component, which leads to artifacts on smooth gradients.
- Compression efficiency. Jpegli compresses images more efficiently than traditional encoders, which saves resources.
How Jpegli works
Jpegli uses a number of new techniques to reduce noise and improve image quality. These are:
- adaptive quantization heuristic for JPEG XL;
- improved selection of the quantization matrix;
- accurate calculation of intermediate results;
- the ability to use a more advanced color space.
Google conducted a test comparing the performance of Jpegli, libjpeg-turbo, and MozJPEG at several bitrates. You can see the results in the diagram below.