Raster-graphics file format supporting lossy and lossless compression
JPEG XL is a royalty-free raster-graphics file format that supports both lossy and lossless compression. It is designed to outperform existing raster formats and thus to become their universal replacement.[2]
Features
The main features are:[4][5]
- Improved functionality and efficiency compared to traditional image formats (e.g. JPEG, GIF and PNG);
- Image dimensions of over a billion (230−1) pixels on each side;[6]
- Up to 4099 channels: 1 grayscale or 3 RGB channels, and optionally up to 4096 "extra" channels like alpha, depth or thermal data;[6]
- Progressive decoding (by resolution and precision);
- Lossless JPEG transcoding with ~20% size reduction;
- CMYK JPEGs are not supported for transcoding, but they are very rare;[7]
- Transcoding to progressive JPEG XLs is supported by the format but not yet implemented in the reference software;[8]
- Lossless encoding and lossless alpha encoding;
- Support for both photographic and synthetic imagery;
- Graceful quality degradation across a large range of bitrates;
- Perceptually optimized reference encoder;
- Support for wide color gamut and HDR;
- Support for animated content,
- Efficient encoding and decoding without requiring specialized hardware
- In particular, JPEG XL is about as fast to encode and decode as old JPEG using libjpeg-turbo and an order of magnitude faster to encode and decode compared to HEIC with x265.[6] It is also parallelizable.
- Royalty-free format with an open-source reference implementation.[9]
Technical details
JPEG XL codec architecture diagram
JPEG XL is based on ideas from Google's PIK format and Cloudinary's FUIF format (which was in turn based on FLIF).[10]
The format is mainly based on two encoding modes:
- VarDCT (variable-blocksize DCT) – utilizes the same DCT algorithm as legacy JPEG, but blocks instead of being restricted to 8x8 come in various sizes (2x2 up to 256x256), non-square shapes (e.g. 16x8, 8x32, 32x64) or can use another transforms (AFV, Hornuss).
- Modular – responsible, among others, for efficient lossless content encoding. This is the mode used for saving adaptive quantization fields, additional channels (e.g. alpha, depth, thermal, spot colors etc.) and DC coefficients (1:8 downsampled image) of the VarDCT mode. It also allows lossy compression with the help of modified Haar transform (called "squeeze"), which has progressive properties: the quality of the image increases with the amount of data loaded. One of the ways VarDCT-based images can be loaded progressively is by saving VarDCT's DC coefficients with Modular's "squeeze", making both modes work in tandem.
Both modes can be assisted by separate modeling of specific image features, unknown in other codecs at the time of creating the format:[a]
- splines for coding e.g. hairs,
- repeating "patches" like text, dots or sprites,
- noise synthesis (since noise is hard to encode in both main modes, it's better to estimate its value in the encoder, save it, and then regenerate the noise in the decoder).
Lossy modes typically use the XYB color space derived from LMS.[12]
JPEG XL can losslessly reencode existing JPEG files by directly copying JPEG's DCT block coefficients to 8x8 VarDCT blocks, making smaller filesizes possible due to better entropy coding. JPEG reconstruction data allows transcoding back to the original JPEG file, although constraints limit support for some files.[13]
Prediction is run using a pixel-by-pixel decorrelator without side information, including a parametrized self-correcting weighted ensemble of predictors. Context modeling includes specialized static models and powerful meta-adaptive models that take local error into account, with a signaled tree structure and predictor selection per context. Entropy coding is LZ77-enabled and can use both asymmetric numeral systems (ANS) and Huffman coding (for low complexity encoders or for reducing overhead of short streams).[citation needed]
It defaults to a visually near-lossless setting that still provides good compression.[6]
Animated (multi-frame) images do not perform advanced inter-frame prediction, though some rudimentary inter-frame coding tools are available:
- frames can update only parts of the canvas;
- besides replacing parts of the canvas, frames can also be blended, added or multiplied to parts of it;[14]
- up to four frames[15] can be remembered and referenced using the "patches" coding tool in later frames.[16]
History
In 2017, JTC1/SC29/WG1 (JPEG) issued a call for proposals for JPEG XL – the next generation image coding standard.[32]
The file format (bitstream) was frozen on 25 December 2020[33] and formally standardized on 13 October 2021.[34] Any file compliant to this standard is guaranteed to be decodable by all future releases.
