Google has introduced a new open source library, jpegli, that implements a JPEG image encoder and decoder. The library includes additional optimizations to improve encoding efficiency, allowing for up to 35% higher compression rates for high-quality images than traditional JPEG codecs. Compared to libjpeg-turbo, the jpegli library allows you to achieve a similar level of quality while reducing the bitrate by 32%. At the API and ABI level, the library is fully compatible with libjpeg62 and can be used to transparently replace it. The library code is written in C++ and distributed under the BSD license.
Increasing the level of compression is achieved by using advanced technologies to reduce image noise and increase quality, using more effective psycho-visual modeling methods to reduce the resulting artifacts. In particular, jpegli uses the adaptive quantization heuristics used by the JPEG XL project, as well as improved algorithms for selecting quantization matrices and calculating intermediate results.
The jpegli library allows you to encode images with 10 or more bits per color component. At the same time, the result of the new encoding algorithms is adapted to the traditional model for the JPEG format, which allows the use of only 8 bits per color component. This feature allows you to maintain compatibility with existing decoders designed for 8-bit representation of color components.
Images encoded using jpegli fully comply with the JPEG standard, do not require specific decoders and can be viewed in existing JPEG viewers and web browsers. Using your own decoder to decompress images compressed using jpegli allows you to further reduce artifacts. Encoding speed using jpegli is comparable to the libjpeg-turbo and MozJPEG libraries.
Source: opennet.ru
