Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

The JPEG XL Image Coding System: History, Features, Coding Tools, Design Rationale, and Future

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read JPEG XL is a royalty-free, universal image codec designed to replace JPEG, PNG, GIF, WebP, AVIF, and HEIC, with roughly 50% smaller lossy files than JPEG and lossless JPEG recompression at about 20% savings.

desk verdict A definitive technical reference for JPEG XL, but the 50% lossy savings claim sits in a benchmark section that is not auditable in the current version. read the letter →

arxiv 2506.05987 v2 pith:NGUXIPR4 submitted 2025-06-06 cs.MM

Jon Sneyers , Jyrki Alakuijala , Luca Versari , Zoltán Szabadka , Sami Boukortt , Amnon Cohen-Tidhar , Moritz Firsching , Evgenii Kliuchnikov
show 4 more authors
Tal Lev-Ami Eric Portis Thomas Richter Osamu Watanabe
This is my paper · ORCID
classification cs.MM
keywords JPEGXLimagecodeclosslesscompressionlossyperceptualqualityrecompressionXYBcolorspaceprogressivedecoding
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

JPEG XL is a royalty-free image coding system published as a multi-part international standard that aims to be a single universal replacement for existing raster image formats. The article argues that JPEG XL matches or outperforms prior formats in both functionality and compression, with lossy and lossless coding across a broad quality range and roughly half the storage and bandwidth cost of JPEG at comparable visual quality. Its distinctive feature is lossless recompression of existing JPEG files, shrinking them by about 20% while allowing byte-for-byte reconstruction of the original JPEG. The paper documents the history, design rationale, coding tools, and future potential of the format, serving as a companion to the standard. The core truth the authors are trying to establish is that one codec can cover the whole still-image space without format selection or migration losses.

What carries the argument

The load-bearing mechanism is the codestream architecture that unifies two coding modes. VarDCT mode, for lossy photographic data, applies variable-size DCT blocks from 8-by-8 to 256-by-256, codes a 1:8 low-frequency image that doubles as a progressive preview, and uses adaptive quantization, chroma-from-luma prediction, and local restoration filters. Modular mode, for lossless data, auxiliary channels, and VarDCT control signals, uses reversible color transforms, palette and delta-palette transforms, the Squeeze transform (a nonlinear Haar-like decomposition), and meta-adaptive decision trees that choose a predictor and an entropy context for each sample. Both modes feed a single hybrid entropy coder that combines prefix coding, asymmetric numeral systems (ANS), and LZ77 dictionary matching, and the bitstream is organized into independently decodable groups. This superset design is what lets JPEG XL absorb the coding tools of JPEG, PNG, GIF, and WebP and add the small reconstruction-data payload that makes original JPEG files recoverable byte-for-byte.

What would settle it

Take a broad, independent corpus of JPEG files—including progressive, arithmetic-coded, and CMYK variants—transcode them with the reference implementation, and check both the fraction that reconstruct byte-for-byte with the reconstruction payload and the average size reduction; then run a blind pairwise visual-quality comparison against optimized JPEG, AVIF, WebP, and HEIC encoders at matched file sizes. If typical JPEGs fail to reconstruct or the visual-quality-at-size advantage is far below the claimed ~50%, the central claim is empirically false.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that JPEG XL is the first image coding system that can serve as a universal interchange format: a superset of the functionality of JPEG, PNG, GIF, WebP, AVIF, HEIC, TIFF, and OpenEXR, with compression performance that is state-of-the-art in both lossless and lossy regimes. The decisive assertions are three: lossy files about 50% smaller than JPEG at matched visual quality across a broad quality range; lossless compression that improves on PNG for all content types; and lossless JPEG recompression with about 20% file-size reduction plus bit-exact recovery of the original JPEG. The paper presents this as a consequence of design choices—two coding modes (VarDCT and Modular), a perceptually motivated internal color space, a flexible progressive bitstream, and a superset of JPEG's coding tools—rather than as an incremental tweak to any one existing format.

Load-bearing premise

The headline numbers rest on a fair comparison: "state-of-the-art" means beating well-tuned competing encoders at matched visual quality, judged by perceptual metrics that are not tilted toward JPEG XL's own error distribution.

Editorial extensions

If this is right

  • Existing JPEG archives can be migrated to JPEG XL at about 80% of their original size, with the original JPEG recoverable byte-for-byte, so no generation loss is introduced.
  • A single format could cover authoring, storage, interchange, and web delivery for photographic and non-photographic raster images, removing the need to choose among JPEG, PNG, GIF, and WebP.
  • Lossy photographic delivery would need roughly half the bandwidth of JPEG at matched visual quality, including high-dynamic-range and wide-gamut content.
  • Lossless compression would improve on PNG for screenshots, pixel art, and other non-photographic images, while the mandatory progressive baseline in VarDCT mode would make previews and region-of-interest decoding standard features.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper leaves implicit that, if the ~20% lossless JPEG recompression figure holds on real-world corpora, archives could store JPEG XL files plus reconstruction data and regenerate the original JPEGs on demand, making the format a transparent archival layer rather than a replacement.
  • A testable extension beyond the paper's reported benchmarks is to stress the universality claim on pathological inputs—bi-level scans, huge flat-color regions, and high-bit-depth floating-point images—where the Modular palette and Squeeze transforms either shine or fail visibly.
  • The same dual-mode architecture suggests a natural next experiment: measuring how far the Modular/VarDCT split can be pushed toward mixed document coding, where text-heavy regions use Modular's palettes and photographic regions use VarDCT, in a single bitstream.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper presents a comprehensive technical overview of the JPEG XL image coding system, covering its history, architecture, file format, color spaces (including XYB), the Modular and VarDCT coding modes, entropy coding, and bitstream ordering. It is written as a companion to the ISO/IEC 18181 standard. The central claims are that JPEG XL provides state-of-the-art compression performance, cutting storage and bandwidth costs in half compared to JPEG for both lossy and lossless compression, and that it can losslessly recompress existing JPEG files with about 20% size reduction and bit-exact reconstruction. The text available for review ends at Section 6.4.1, so the compression-performance evidence (Section 10) is not included.

Significance. If the claims hold, JPEG XL would be a significant advance in image compression, and this paper would be the most complete open reference to the standard. The descriptive sections (1–9) are detailed, internally consistent, and agree with the public standard and libjxl behavior. The paper is commendably honest about implementation limitations, including the 24-bit integer precision cap in libjxl (Section 4.1.4), the intentional lack of lossless JPEG recompression for CMYK (Section 4.3.5), and the restriction to 2× chroma subsampling factors (Section 4.2.3). The main weakness is that the empirical claims that motivate the paper are not auditable from the text provided: Section 10 is absent, and the lossy quality metric is authored by the same team, so the central '50% saving' claim needs a transparent benchmark to be convincing.

major comments (3)
  1. [TL;DR; Section 10] The TL;DR (p. 2) and the 'compression' bullet list state that JPEG XL 'cut[s] storage and bandwidth costs in half compared to JPEG' and that lossless JPEG recompression reduces file sizes 'by about 20%.' These are empirical claims that require a benchmark protocol. The text provided for review ends at Section 6.4.1 (Quantization tables), and Section 10 (Compression performance), although listed in the contents, is not present. As a result, the reader cannot audit the corpus, the encoder versions and settings, the quality-matching procedure, or the metrics underlying the headline numbers. Please include the full Section 10 in the version sent to reviewers, with a specification of the image corpus, the exact encoder configurations (e.g., libjxl effort level, libjpeg-turbo quality, AVIF/WebP/HEIC settings), and the quality-matching method; if the numbers are instead drawn from published third-party evaluations, cite them explicitly.
  2. [§1.1.12, §4.2.1] Butteraugli and the XYB color space are described in Sections 1.1.12 and 4.2.1 as having been designed by the authors themselves. The lossy '50% saving' claim is likely to be evaluated with Butteraugli (or a metric strongly correlated with it), and XYB is the internal color space that Butteraugli was designed to judge. If the quality levels of competing codecs in Section 10 are matched using Butteraugli, the comparison is not neutral: it may favor JPEG XL's error distribution over that of JPEG, AVIF, WebP, or HEIC. The paper should explicitly state which metric(s) govern the quality matching in Section 10, and ideally report the main comparison also under an independent metric (e.g., SSIM, VMAF, or a formal subjective test). This is not an accusation of bias; it is a request for the methodological transparency needed to make the lossy claim falsifiable.
  3. [§1.1.7, §5] The lossless compression claim ('substantial lossless compression improvements over PNG regardless of the type of image content', Section 1.1.7; 'saving about 50%, both for lossy and lossless', TL;DR) is also a comparative claim. The lossless case is a file-size fact and inherently less metric-dependent, but the claim's scope ('regardless of the type of image content') still requires a defined corpus covering photographic, synthetic, and mixed content. The Modular-mode description in Section 5 is detailed and plausible, but without the Section 10 data the universality of the lossless improvement cannot be verified.
minor comments (5)
  1. [§1.1.3] In the discussion of the de facto JPEG format, the phrase 'ade factostandard' appears without spaces; it should read 'a de facto standard.'
  2. [§1.1.4] PNG is officially 'Portable Network Graphics' (plural); the text reads 'Portable Network Graphic.'
  3. [§2.4.6] In Section 2.4.6, the level limits are given as '228 (268 megapixels)' and '240 pixels (1099 gigapixels)'; the superscripts are lost in the extraction and should read 2^28 and 2^40.
  4. [Figure 24] In the property list of Figure 24, property index 23 appears twice (once as PrevChannelErr and once as |Prev3Channel|); the enumeration should be renumbered.
  5. [§1.2.1] The phrase 'Draft Call for Proposals 1' contains a stray footnote marker '1' that is not defined; please remove or explain it.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the paper is a descriptive codec overview whose compression claims are empirical and not reduced by construction to the authors' own metric or color space.

full rationale

The reviewed text is a technical survey of JPEG XL and its coding tools, not a derivation chain whose outputs are defined by its inputs. The headline claims—lossless recompression saving about 20% and lossy compression cutting costs in half compared to JPEG—are empirical performance statements. The lossless recompression claim is tied to entropy-coding improvements over JPEG's Huffman coding, which is a concrete coding mechanism rather than a tautology. The lossy claim depends on Section 10, which is not present in the supplied text; without that benchmark protocol, the claim cannot be audited, but absence of evidence is not circularity. The paper does use Butteraugli, a metric designed by the same authors, and XYB, their own color space, as design and evaluation tools (e.g., 'Butteraugli, a psychovisual distortion metric developed in 2016 to facilitate the optimization of high-fidelity image encoders' and 'The XYB color space is derived from perceptual models [19] based on the absorbance spectra of the long (L), medium (M), and short (S) wavelength cone cells'). However, no equation or construction in the text makes a predicted file size equal to a fitted parameter or to a Butteraugli score. The metric and color space influence encoder decisions, but the resulting bitstream sizes are independently measurable facts. Self-citations to ISO/IEC 18181 and earlier overviews are contextual and not load-bearing for a derivation. A benchmark that uses the authors' own metric as the sole quality yardstick would be a methodology concern, but the paper does not, within the reviewed text, define its central claim in terms of that metric. Therefore no specific circular reduction can be exhibited, and the appropriate finding is no significant circularity.

Assumptions & free parameters 5 free parameters · 4 assumptions · 2 invented entities

The central performance claims rest on a set of design constants (XYB transform, quantization tables, filter weights) that are stated in the paper but fitted during the codec's development by the authors' own perceptual research. The axioms are the perceptual assumptions that justify XYB and metric-based quality comparisons. The invented entities are XYB itself and the Butteraugli metric used to evaluate it.

free parameters (5)
  • XYB LMS matrix and bias b = b = 0.00379307325527544933; matrix rows 0.3/0.622/0.078, 0.23/0.692/0.078, 0.2434227/0.2047674/0.5518099 (Eq. 1)
    These constants fix the absolute XYB color space that underlies all lossy coding; they are design choices from the authors' psychovisual research, stated without derivation from independently published cone-response data.
  • Cube-root transfer exponent = 1/3 (Eq. 2)
    Chosen for perceptual behavior and for cheap decoder implementation (cube in the inverse); no uniqueness theorem forces this exponent.
  • Default Gaborish filter weights = 0.115169525 and 0.061248592
    Hand-tuned defaults for the 3x3 deblocking filter, signaled in the frame header as gab_x_weight and gab_y_weight.
  • EPF default parameters = sigma 0.46, sigma scales 6.5 and 2/3, channel scales 40.0/5.0/3.5
    Hand-tuned defaults for the edge-preserving filter; these directly control artifact suppression at low bitrates and hence the quality comparisons.
  • Default X/Y/B quantization tables = Figure 31 tables (relative per-component values)
    Hand-designed quantization tables for DCT8x8 per XYB component; they directly determine lossy rate-distortion and are stated in the paper without independent derivation.
assumptions (4)
  • domain assumption S-cone contribution to perceived luminance is negligible at high spatial frequencies but substantial at large angular scales
    Drives the XYB design (Section 4.2.1, Motivation). If false, the XYB noise allocation and the lossy quality claims weaken.
  • domain assumption Perceptual metrics (Butteraugli and comparable metrics) rank codec output the way human observers would
    The lossy comparisons in the paper treat metric quality as visual quality (Sections 1.1.12 and 10.2).
  • domain assumption Foveal S-cone density is low enough to justify down-weighting B-contrast at high frequencies
    Invoked via citation [18] in Section 4.2.1 to explain the foveal S-cone sparsity.
  • standard math Standard mathematical tools (DCT, ANS, non-linear Haar/Squeeze) behave as documented
    Background assumptions for Sections 5, 6, and 8; these are well-established techniques.
invented entities (2)
  • XYB color space independent evidence
    purpose: Absolute, perceptually uniform internal color space for all lossy coding; anchors quantization and quality decisions
    Fully specified by the equations in Section 4.2.1 and public in libjxl; the perceptual benefit is falsifiable through independent human-subject comparisons, though no such external test is presented here.
  • Butteraugli perceptual metric independent evidence
    purpose: Psychovisual distortion metric used as the quality yardstick in lossy evaluations and encoder optimization
    Public implementation and published description; the metric's correlation with human judgment is testable, but its use as the yardstick for the codec's own claims is self-referential (Section 1.1.12).

how reviews work

0 comments
Cite this review

Pith. "Pith review of The JPEG XL Image Coding System: History, Features, Coding Tools, Design Rationale, and Future." pith.science (2026). https://pith.science/paper/NGUXIPR4

@misc{pith2026250605987,
  author       = {Pith},
  title        = {Pith review of: The JPEG XL Image Coding System: History, Features, Coding Tools, Design Rationale, and Future},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NGUXIPR4}},
  note         = {Machine review of arXiv:2506.05987}
}
read the original abstract

JPEG XL is a new image coding system offering state-of-the-art compression performance, lossless JPEG recompression, and advanced features. It aims to replace JPEG, PNG, GIF, and other formats with a single universal codec. This article provides an overview of JPEG XL, including its history, design rationale, coding tools, and future potential. It can be used as a companion document to the standard (ISO/IEC 18181), or as a standalone article to better understand JPEG XL, either at a high level or in considerable technical detail.

Figures

Figures reproduced from arXiv: 2506.05987 by the authors.

Figure 1
Figure 1. Schematic illustration of the scope of the JPEG XL standard and its positioning with respect to other image formats. The horizontal axis depicts a spectrum from purely photographic raster images (left) to purely synthetic vector graphics (right), with a range in between that contains mixed content and non-photographic raster images. The vertical axis depicts various stages of the digital asset lifecycle, from captur… view at source ↗
Figure 2
Figure 2. Evolution of camera and display resolutions. way. Microsoft’s RIFF container was inspired by IFF, and so was Apple’s QuickTime container format, which would become the basis for ISOBMFF (e.g. MP4 and HEIF). Also arguably the PNG chunk structure was inspired by IFF. The Interleaved Bitmap (ILBM) format, popular on the Commodore Amiga platform, was based on IFF. It offered palette images with up to 256 colors (with la… view at source ↗
Figure 3
Figure 3. Timeline of image formats and their adoption on the Web. option (which can be useful in case very fast encoding is required) but there are also better options available, includ￾ing Asymmetric Numeral Systems (ANS) and an optional use of LZ77. Additionally, JPEG XL adds sophisticated context modeling options, which further improves entropy coding results. Since the entropy coding step in both JPEG and JPEG XL is itse… view at source ↗
Figures from the paper (55 more)
Figure 4
Figure 4. Figure 4: Objective metrics can give deceiving results when an image has an ‘easy’ background. The best (highest quality) metric score is indicated in bold. The image on the left has a better visual quality than the image on the right, which has a large amount of padding causing…
Figure 5
Figure 5. Figure 5: Overview of various containers, payload codecs, and image file formats. ated a different, more generic image container format called HEIF [24] which became an MPEG standard (ISO/IEC 23008-12) in 2015. The HEIF container is based on the ISO base media file format (ISOBM…
Figure 6
Figure 6. Figure 6: Overview of the JPEG XL codec architecture [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Image header syntax. Note that most fields are signaled conditionally and have a variable length [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Frame header syntax [PITH_FULL_IMAGE:figures/full_fig_p021_8.png]
Figure 9
Figure 9. Figure 9: Spectral sensitivity of human cone cells [57]. medium (M), and short (S) wavelength cone cells of the human retina. Structurally analogous to opponent-process color spaces like CIELAB, XYB utilizes three axes: a lumi￾nance channel (Y) representing black to white, a red…
Figure 10
Figure 10. Figure 10: Illustration of the distribution of cone cells in the fovea, from [18]. b = 0.00379307325527544933 Lm = 0.3Rl +0.622Gl +0.078Bl +b (1) Mm = 0.23Rl +0.692Gl +0.078Bl +b Sm = 0.2434227Rl +0.2047674Gl +0.5518099Bl +b Compared to the Hunt-Pointer-Estevez definition of ´ L…
Figure 11
Figure 11. Figure 11: Comparison between various transfer functions. X = (Lg −Mg)/2 (3) Y = (Lg +Mg)/2 B = Sg While B corresponds to just the S cone response, in practice in JPEG XL it will rarely be used directly. In VarDCT mode, ‘chroma from luma’ is typically used to subtract Y from B, …
Figure 12
Figure 12. Figure 12: Visualization of the XYB’ color space: slices of the XB’ plane at four values of Y. have a device pixel ratio (DPR) of 2 or even 3, which cor￾responds to a pixel size of about 0.01°, i.e. more than two orders of magnitude smaller than the stimuli typically used to det…
Figure 14
Figure 14. Figure 14: Slices of the XB’ plane at four values of Y, with the color gamuts of sRGB, P3, and BT.2020 indicated. d alpha (Bool): This field signals a particularly common set of defaults. If true, the extra channel is of type Al￾pha (unassociated), 8-bit, at 1:1 resolution and w…
Figure 15
Figure 15. Figure 15: Visualization of the YCoCg-R color space: slices of the CoCg plane at four values of Y. of (Co,Cg) colors, 2) a 256 × 256 luma (Y) channel, 3) a 256 × 256 index channel with samples in range [0,99] that refer to the palette, and finally, 4) a 256 × 256 alpha channel. …
Figure 16
Figure 16. Figure 16: Decomposition of an example image into color components. Top: color image (left) and RGB components, which are obviously highly correlated. Middle: YCoCg-R, a decorrelation transform suitable for lossless compression. Bottom: XYB, a perceptually motivated color decorr…
Figure 17
Figure 17. Figure 17: Left: the first 64 colors of the implicit palette, forming a 4×4×4 cube in the RGB space of the image (here assumed to be sRGB). Right: the next 125 colors of the implicit palette, forming a 5×5×5 cube (inner 3×3×3 colors omitted for clarity). index color hexcode inde…
Figure 19
Figure 19. Figure 19: The implicit delta palette entries. Each delta entry corresponds to a difference vector that gets added to the predicted color; the colors shown here correspond to the resulting color if the predicted color is gray. a c A B b 1.0 0.5 0.5 1.5 [PITH_FULL_IMAGE:figures/…
Figure 18
Figure 18. Figure 18: The implicit palette, in case of 8-bit sRGB. 150 100 50 0 50 100 150 R-delta 150 100 50 0 50 100 150 G-delta 150 100 50 0 50 100 150 B-delta [PITH_FULL_IMAGE:figures/full_fig_p031_18.png]
Figure 21
Figure 21. Figure 21: Top: result of the forward Squeeze transform with default parameters: a series of residual channels of increasing dimensions. Bottom: intermediate results and final result of applying the inverse Squeeze transform. linear interpolation tendency reconstruction accordin…
Figure 22
Figure 22. Figure 22: Examples illustrating the difference between unconditional linear interpolation and the nonlinear tendency term [PITH_FULL_IMAGE:figures/full_fig_p033_22.png]
Figure 23
Figure 23. Figure 23: Crop of an image compressed to 0.4 bpp using the Squeeze transform with quantization of the residuals. From left to right: 1) no tendency term at all, 2) an unconditional linear interpolation tendency term, 3) the nonlinear tendency term used in JPEG XL, 4) original i…
Figure 24
Figure 24. Figure 24: lists all the local properties that can be refer￾enced in an MA tree decision node. Some of these properties are static within the inner cod￾ing loop, which means that in an optimized implementation, it can be worthwhile to specialize the tree to avoid unnec￾essary br…
Figure 25
Figure 25. Figure 25: Predictors available in Modular coding. Division is integer division, rounding towards zero. with the same dimensions as the current channel. PrevChan￾nelErr corresponds to the prediction miss w.r.t. the clamped gradient predictor. Prev2Channel is the same as PrevChan…
Figure 26
Figure 26. Figure 26: Example of how the various predictors would continue an initial image pattern if all prediction residuals would be zero. (link to jxl-art) The Weighted (or Self-correcting) predictor is rather complicated. It works particularly well for photographic images. It will be…
Figure 27
Figure 27. Figure 27: Modular encoding at different libjxl effort settings. Efforts 2 and 3 use a fixed predictor and a fixed MA tree; efforts 4 and 7 use two different predictors and a constructed MA tree; effort 11 uses all possible predictors. The leaf nodes of MA tree trees are filled …
Figure 28
Figure 28. Figure 28: The block sizes and types available in VarDCT mode. The number in the bottom-right corner is the index of the block type, used to signal the image segmentation / block type selection. If two block types have the same color, they are rotated variants of the same transf…
Figure 29
Figure 29. Figure 29: Example image (left) and two possible ways to segment it into various block types, corresponding to choices made by the libjxl v0.11 encoder. The middle image corresponds to a relatively high fidelity setting (distance 1), the image on the right to a lower fidelity se…
Figure 30
Figure 30. Figure 30: Color banding artifacts in slow gradients encoded at a low quality setting in different codecs. ally, it can be signaled that an adaptive LF smoothing step has to be applied by a decoder after dequantization. This step has the effect of smoothing the LF image in regio…
Figure 31
Figure 31. Figure 31: Default quantization tables for DCT8x8. In this figure, the factors are rescaled per component to be relative to the DC factor, since the differences in the scales of the X,Y,B components would otherwise make it hard to compare them. The coloring goes from green (fine…
Figure 32
Figure 32. Figure 32: Default quantization tables for DCT32x16 [PITH_FULL_IMAGE:figures/full_fig_p042_32.png]
Figure 33
Figure 33. Figure 33: Heatmap showing adaptive quantization weights, for the middle image of [PITH_FULL_IMAGE:figures/full_fig_p044_33.png]
Figure 34
Figure 34. Figure 34: Four example splines and the corresponding 1D DCT coefficients representing the evolution of the X,Y,B components and Sigma (thickness) along the arclength. (link to jxl-art) Original image Reference frame Residual image [PITH_FULL_IMAGE:figures/full_fig_p046_34.png]
Figure 35
Figure 35. Figure 35: Example usage of the Patches coding tool. Without Patches, the image on the left (297 kB as a PNG file) can be losslessly compressed to a 162 kB JPEG XL file (0.59 bpp). With Patches, this can be reduced to 87 kB (0.32 bpp) by extracting repeated image elements, codin…
Figure 36
Figure 36. Figure 36: Luma-modulated noise generation with various parameter settings. From top to bottom: no noise, photon noise corresponding to ISO 400, ISO 6400, and ISO 1600, and at the bottom noise parameters where the dark regions have no noise while the bright regions do (lookup ta…
Figure 37
Figure 37. Figure 37: Zoomed crop of an image to demonstrate the effect of the Gabor-like transform Top: original image. Middle: compressed image at 0.73 bpp, Gaborish disabled. Note the visible DCT macroblocks. Bottom: compressed image at the same bit rate (0.73 bpp), with Gaborish enable…
Figure 38
Figure 38. Figure 38: Zoomed crop of an image to demonstrate the effect of the edge-preserving filter (EPF). Left: original image. Middle: compressed image, with a relatively low fidelity setting (distance 4), EPF disabled. Note the DCT ringing artifacts, as well as some DCT base patterns …
Figure 39
Figure 39. Figure 39: Weights for the non-separable 8x upsampling method. Left: indices of the weights. Only 210 of the 1600 weights are different and are optionally signaled. Right: default weights. Weights in the figure are scaled by a factor of 100. the input sample. The method is non-s…
Figure 40
Figure 40. Figure 40: Result of 8x upsampling using different methods. Cubic and Lanczos upsampling are as implemented in ImageMagick -resize with respectively -filter cubic and -filter lanczos [PITH_FULL_IMAGE:figures/full_fig_p050_40.png]
Figure 41
Figure 41. Figure 41: Custom weights for 8x upsampling: a variation on nearest-neighbor. Left: weights. Right: example result. 8.1 Hybrid integer coding The configuration of hybrid integer coding has three param￾eters: split exponent, msb in token, and lsb in token. Arbitrary unsigned 32-b…
Figure 43
Figure 43. Figure 43: Special LZ77 distance codes for Modular data. 8.1.3 ANS Asymmetric Numeral Systems [16] is a generic entropy coding method that combines the fast decode speed of prefix coding and the precise probability distributions of arithmetic coding, leading to better compressio…
Figure 42
Figure 42. Figure 42: Hybrid integer coding: example configurations. The configurations are specified as a triple of the form split exponent, msb in token, lsb in token. 119 116 111 106 97 88 84 74 72 75 85 89 98 107 112 117 118 113 103 92 80 68 60 56 54 57 61 69 81 93 104 114 115 108 94 7…
Figure 44
Figure 44. Figure 44: Context clustering example. The histograms corresponding to the 104 pre-clustering contexts (left) are clustered and mapped into 27 post-clustering contexts (right), reducing the total signaling cost. There are two caveats: 1) LZ77 refers to previously coded symbols, …
Figure 45
Figure 45. Figure 45: Segmentation of an image frame into groups. JPEG XL file (ISO/IEC 18181-2) JPEG XL codestream (ISO/IEC 18181-1) n JXL␣ ftyp jxli Exif jbrd jxlc 0xFF0A brobxml␣ jumb Signature File type Frame index Exif metadata JPEG bitstream reconstruction data JPEG XL codestream box…
Figure 46
Figure 46. Figure 46: Example bitstream structure of a JPEG XL file and the codestream it contains. There is one exception to this general section layout: if the frame is small and there is only a single HF group — in other words, the frame dimensions are at most 256×256, in case of VarDCT…
Figure 47
Figure 47. Figure 47: Progressive preview after loading 800 kB of a 2,274 kB JPEG XL file (35%), with default ordering and with center-first permuted ordering. Photo by Alexandru Zdrobau on Unsplash. ˘ Default section order: Permuted section order: Permutation: Lehmer code: 6 7 8 9 10 11 1…
Figure 48
Figure 48. Figure 48: Default ordering and center-first permuted section ordering. HF group section numbers start at 6 since indices 0-5 are used for LfGlobal, LF Groups and HfGlobal. Note how the Lehmer code reduces the amplitude of the numbers in the sequence to be entropy coded in permu…
Figure 49
Figure 49. Figure 49: Error resilience demonstration. Left: original image, with an indication of four HF group regions for which corresponding JPEG XL bitstream sections are corrupted. Right: image that can be recovered from the corrupted bitstream. 9.2.3 LF frames The LF coefficients eff…
Figure 50
Figure 50. Figure 50: Example of a layered image with two smaller layers alpha-blended over a base layer. Original photo by Talha Resitoglu on Pexels. handle a single decoded image frame will correctly render the merged image. The layer separation is preserved, in￾cluding an optional name …
Figure 51
Figure 51. Figure 51: Example of reversible cropping by adjusting the image canvas and frame offset. 10. Compression performance In general, it is not possible to evaluate the ‘inherent’ per￾formance of a codec, in terms of speed, compression or quality. It is only possible to test specifi…
Figure 52
Figure 52. Figure 52: Lossless compression results for four sets of images. Pareto-optimal trade-offs between encode speed and compression density are underlined and have a blue background; best results are indicated in bold. Results that would be Pareto-optimal if not for JPEG XL are indi…
Figure 53
Figure 53. Figure 53: Lossless JPEG recompression results. The test image corpus is the set of 8-bit color photos from imagecompression.info/test images. They are compressed with various JPEG encoders and quality settings, and then recompressed using libjxl v0.11.1 with default settings. J…
Figure 54
Figure 54. Figure 54: Lossy compression performance of JPEG XL compared to JPEG and AVIF [61]. The vertical axis indicates distortion in JND units, as estimated using the AIC-3 methodology using boosted and plain triplet comparisons. The shaded regions indicate a 95% confidence interval. L…
Figure 55
Figure 55. Figure 55: Aggregated bitrate-distortion plots for a set of 49 SDR images (subset1) according to CVVDP [43]. Every data point corresponds to an encoder quality setting (sampled from 1 to 99 in steps of 2), with the average bit rate on the horizontal axis and the aggregated metri…
Figure 56
Figure 56. Figure 56: Example of a libjpeg-turbo q19 image, with CVVDP score 9.0634. The image suffers from obvious color banding and blocking artifacts. 0.2 0.29 0.42 0.62 0.91 1.33 1.95 2.85 4.18 6.12 Average bpp for that encoder setting 0 10 20 30 40 50 60 70 80 90 95 Average SSIMULACRA…
Figure 58
Figure 58. Figure 58: Lossy compression performance of JPEG XL compared to JPEG XT and AVIF [36]. The thumbnail images were tone mapped to SDR. The vertical axis indicates distortion in JND units, as estimated using the AIC-3 methodology using boosted and plain triplet comparisons. The sha…
Figure 60
Figure 60. Figure 60: Aggregated results according to CVVDP [43] for a set of 99 images, of which 24 are the HDR images from [PITH_FULL_IMAGE:figures/full_fig_p066_60.png]
Figure 61
Figure 61. Figure 61: Encode/decode speed measurements for various encoder settings. consistency than others. The codec design itself does play a role too though: e.g. if there are restoration filters available but the bitstream does not allow fine-grained signaling of filtering strengths,…
Figure 62
Figure 62. Figure 62: Percentage of image requests coming from a browser with JPEG XL support, as estimated by Cloudinary based on aggregate statistics from approximately 15 billion images served per day. 2025, across all Microsoft platforms: Xbox console, PC, Mobile, Surface Hub, and Holo…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. JPEG AIC2026: A large-scale dataset for fine-grained assessment of image coding

    eess.IV 2026-07 conditional novelty 7.0 of 10

    AIC2026 is a large-scale benchmark of 9,618 fine-grained distorted images spanning 17 conventional and learned codec configurations across 20 CVVDP-based perceptual levels.

  2. Proteus: A Truncation-Robust Entropy Model for Progressive LiDAR Compression

    cs.CV 2026-08 conditional novelty 6.0 of 10

    A LiDAR codec that keeps the most significant range bits in a self-contained stream and encodes the rest in a FIFO stream, making any prefix of the truncatable stream decode to a deterministically coarser point cloud.

Reference graph

Works this paper leans on

71 extracted references · 67 canonical work pages · cited by 2 Pith papers

  1. [1]

    Digital Negative (DNG) specification version 1.7.1.0, September 2023

    Adobe. Digital Negative (DNG) specification version 1.7.1.0, September 2023. https://helpx.adobe. com/camera-raw/digital-negative.html

    work page 2023

  2. [2]

    Assessment of quality of JPEG XL proposals based on subjective methodologies and objective metrics

    Pinar Akyazi and Touradj Ebrahimi. Assessment of quality of JPEG XL proposals based on subjective methodologies and objective metrics. InApplica- tions of Digital Image Processing XLII, volume 11137, pages 147–163. SPIE, 2019

    work page 2019

  3. [3]

    Jyrki Alakuijala and Zolt ´an Sz- abadka

    IETF RFC 7932. Jyrki Alakuijala and Zolt ´an Sz- abadka. Brotli compressed data format, 2016

    work page 2016

  4. [4]

    Users prefer Guetzli JPEG over same-sized libjpeg

    Jyrki Alakuijala, Robert Obryk, Zolt´an Szabadka, and Jan Wassenberg. Users prefer Guetzli JPEG over same- sized libjpeg.arXiv:1703.04416, 2017

    work page Pith review arXiv 2017

  5. [5]

    JPEG XL next-generation image compression architecture and coding tools

    Jyrki Alakuijala, Ruud Van Asseldonk, Sami Boukortt, Martin Bruse, Iulia-Maria Coms,a, Moritz Firsching, Thomas Fischbacher, Evgenii Kliuchnikov, Sebastian Gomez, Robert Obryk, et al. JPEG XL next-generation image compression architecture and coding tools. In Applications of digital image processing XLII, volume 11137, pages 112–124. SPIE, 2019

    work page 2019

  6. [6]

    Apple’s secret upgrade to the iPhone 16 Pro yields incredible results, October 2024

    Alex Armitage. Apple’s secret upgrade to the iPhone 16 Pro yields incredible results, October 2024. Fstoppers website

    work page 2024

  7. [7]

    Image coding comparisons, Decem- ber 2022

    A VIF team. Image coding comparisons, Decem- ber 2022. https://storage.googleapis.com/ avif-comparison/index.html

    work page 2022

  8. [8]

    Martini, and Yuriy Reznik

    Nabajeet Barman, Maria G. Martini, and Yuriy Reznik. Bjøntegaard Delta (BD): A tutorial overview of the metric, evolution, challenges, and recommendations. arXiv:2401.04039, 2024

    arXiv 2024

Show all 71 references
  1. [9]

    Benchmarking lossless still image codecs: Perspectives on selected compression stan- dards from 1992 through 2022

    Michael J Bennett. Benchmarking lossless still image codecs: Perspectives on selected compression stan- dards from 1992 through 2022. InThe Society for Imaging Science and Technology Archiving 2023 Fi- nal Program and Proceedings, pages 165–171, 2023

    1992

  2. [10]

    Sleator, Robert E

    Jon Louis Bentley, Daniel D. Sleator, Robert E. Tarjan, and Victor K. Wei. A locally adaptive data compres- sion scheme.Commun. ACM, 29(4):320–330, 1986

    1986

  3. [11]

    Users prefer Jpegli over same-sized libjpeg-turbo or MozJPEG.arXiv:2403.18589, 2024

    Martin Bruse, Luca Versari, Zolt ´an Szabadka, and Jyrki Alakuijala. Users prefer Jpegli over same-sized libjpeg-turbo or MozJPEG.arXiv:2403.18589, 2024. The JPEG XL Image Coding System: History, Features, Coding Tools, Design Rationale, and Future — 72/73

    2024 arXiv

  4. [12]

    CIPA DC- 008-Translation-2024 — exchangeable image file for- mat for digital still cameras : Exif version 3.0, 2024

    Camera & Imaging Products Association. CIPA DC- 008-Translation-2024 — exchangeable image file for- mat for digital still cameras : Exif version 3.0, 2024

    2024

  5. [13]

    A class of lo- cal interpolating splines

    Edwin Catmull and Raphael Rom. A class of lo- cal interpolating splines. In Robert E. Barnhill and Richard F. Riesenfeld, editors,Computer Aided Geo- metric Design, pages 317–326. Academic Press, 1974

    1974

  6. [14]

    An overview of coding tools in A V1: the first video codec from the Alliance for Open Media.APSIPA Transactions on Signal and Information Processing, 9, 2020

    Yue Chen, Debargha Mukherjee, Jingning Han, Adrian Grange, Yaowu Xu, Sarah Parker, Cheng Chen, Hui Su, Urvang Joshi, et al. An overview of coding tools in A V1: the first video codec from the Alliance for Open Media.APSIPA Transactions on Signal and Information Processing, 9, 2020

    2020

  7. [15]

    Improvement of JPEG XL lossy image coding us- ing region adaptive DCT block partitioning structure

    Joonhyung Cho, Oh-Jin Kwon, and Seungcheol Choi. Improvement of JPEG XL lossy image coding us- ing region adaptive DCT block partitioning structure. IEEE Access, 9:113213–113225, 2021

    2021

  8. [16]

    Asymmetric numeral systems: entropy coding combining speed of Huffman coding with com- pression rate of arithmetic coding.arXiv:1311.2540, 2014

    Jarek Duda. Asymmetric numeral systems: entropy coding combining speed of Huffman coding with com- pression rate of arithmetic coding.arXiv:1311.2540, 2014

    2014 arXiv

  9. [17]

    The JPEG XR image coding standard [stan- dards in a nutshell].IEEE Signal Processing Maga- zine, 26(6):195–204, 2009

    Fr´ed´eric Dufaux, Gary J Sullivan, and Touradj Ebrahimi. The JPEG XR image coding standard [stan- dards in a nutshell].IEEE Signal Processing Maga- zine, 26(6):195–204, 2009

    2009

  10. [18]

    Fairchild.The Color Curiosity Shop

    Mark D. Fairchild.The Color Curiosity Shop. Rochester Institute of Technology, 2011

    2011

  11. [19]

    Fairchild.Color Appearance Models

    Mark D. Fairchild.Color Appearance Models. Wiley, 3rd edition, 2013

    2013

  12. [20]

    Compression of spectral images using spectral JPEG XL.Journal of Computer Graphics Techniques (JCGT), 14(1):49–69, March 2025

    Alban Fichet and Christoph Peters. Compression of spectral images using spectral JPEG XL.Journal of Computer Graphics Techniques (JCGT), 14(1):49–69, March 2025

    2025

  13. [21]

    Open Source Geospatial Foundation, 2025.https://gdal.org

    GDAL/OGR contributors.GDAL/OGR Geospatial Data Abstraction software Library. Open Source Geospatial Foundation, 2025.https://gdal.org

    2025

  14. [22]

    Objective assessment of the WebP image coding algorithm.Signal processing: image commu- nication, 27(8):867–874, 2012

    Giaime Ginesu, Maurizio Pintus, and Daniele D Giusto. Objective assessment of the WebP image coding algorithm.Signal processing: image commu- nication, 27(8):867–874, 2012

    2012

  15. [23]

    Why Apple uses JPEG XL in the iPhone 16 and what it means for your photos, Septem- ber 2024

    Jeremy Gray. Why Apple uses JPEG XL in the iPhone 16 and what it means for your photos, Septem- ber 2024. PetaPixel website

    2024

  16. [24]

    Hannuksela, Jani Lainema, and Vinod K

    Miska M. Hannuksela, Jani Lainema, and Vinod K. Malamal Vadakital. The High Efficiency Image File Format standard [Standards in a Nutshell].IEEE Sig- nal Processing Magazine, 32(4):150–156, 2015

    2015

  17. [25]

    A method for the construction of minimum-redundancy codes.Proceedings of the Institute of Radio Engineers, 40(9):1098–1101, 1952

    David A Huffman. A method for the construction of minimum-redundancy codes.Proceedings of the Institute of Radio Engineers, 40(9):1098–1101, 1952

    1952

  18. [26]

    A colour- appearance transform for the CIE 1931 standard col- orimetric observer.Color Research & Application, 10(3):165–179, 1985

    Robert WG Hunt and MR Pointer. A colour- appearance transform for the CIE 1931 standard col- orimetric observer.Color Research & Application, 10(3):165–179, 1985

    1931

  19. [27]

    Graphic technology — Prepress digital data exchange — Tag image file format for image technology (TIFF/IT)

    ISO 12639. Graphic technology — Prepress digital data exchange — Tag image file format for image technology (TIFF/IT)

  20. [28]

    Graphic technology — extensible meta- data platform (XMP)

    ISO 16684-1. Graphic technology — extensible meta- data platform (XMP)

  21. [29]

    Information technology — JPEG XL image coding system — Part 1: Core coding system

    ISO/IEC 18181-1. Information technology — JPEG XL image coding system — Part 1: Core coding system

  22. [30]

    Information technology — JPEG XL image coding system — Part 2: File for- mat

    ISO/IEC 18181-2. Information technology — JPEG XL image coding system — Part 2: File for- mat

  23. [31]

    Information technology — JPEG systems — part 5: JPEG universal metadata box for- mat (JUMBF)

    ISO/IEC 19566-5. Information technology — JPEG systems — part 5: JPEG universal metadata box for- mat (JUMBF)

  24. [32]

    Information technology — Assessment of image coding — Part 3: Subjective quality assessment of high-fidelity images

    ISO/IEC DIS 29170-3. Information technology — Assessment of image coding — Part 3: Subjective quality assessment of high-fidelity images

  25. [33]

    JPEG white paper: JPEG XL image coding system

    ISO/IEC JTC 1 / SC29 / WG1 N100400. JPEG white paper: JPEG XL image coding system. 98th JPEG meeting, Sydney, Australia, January

  26. [34]

    JPEG XL use cases and requirements

    ISO/IEC JTC1 / SC29 / WG1 N83043. JPEG XL use cases and requirements. 83rd JPEG Meeting, Geneva, Switzerland, March 2019. https://ds. jpeg.org/documents/wg1n83043-REQ-JPEG_ XL_Use_Cases_and_Requirements.pdf

    2019

  27. [35]

    Coding-independent code points for video signal type identification

    ITU-T recommendation H.273. Coding-independent code points for video signal type identification

  28. [36]

    Lazzarotto, Shima Mohammad, Dominik Keller, Atanas Boev, Ant´onio Pinheiro, Thomas Richter, Alexander Raake, Touradj Ebrahimi, Jo˜ao Ascenso, and Dietmar Saupe

    Mohsen Jenadeleh, Jon Sneyers, Davi N. Lazzarotto, Shima Mohammad, Dominik Keller, Atanas Boev, Ant´onio Pinheiro, Thomas Richter, Alexander Raake, Touradj Ebrahimi, Jo˜ao Ascenso, and Dietmar Saupe. Fine-grained HDR image quality assessment from no- ticeably distorted to very...

    2025

  29. [37]

    C.-A. Laisant. Sur la num ´eration factorielle, ap- plication aux permutations.Bulletin de la Soci ´et´e Math´ematique de France, 16:176–183, 1888

  30. [38]

    MozJPEG 3.0

    Kornel Lesi ´nski. MozJPEG 3.0. Web Per- formance Calendar, December 28th, 2014. https://calendar.perfplanet.com/2014/ mozjpeg-3-0/

    2014

  31. [39]

    Ming Luo, Guihua Cui, and B. Rigg. The devel- opment of the CIE 2000 colour-difference formula: CIEDE2000.Color Research & Application, 26:340 – 350, 10 2001

    2000

  32. [40]

    Maher.The Future Was Here: The Commodore Amiga

    J. Maher.The Future Was Here: The Commodore Amiga. Platform Studies. MIT Press, 2018

    2018

  33. [41]

    A comparative study on lossless compression mode in WebP, better portable graphics (BPG), and JPEG XL image compression algorithms

    Thulfiqar H Mandeel, Muhammad Imran Ahmad, Noor Aldeen A Khalid, and Mohd Nazrin Md Isa. A comparative study on lossless compression mode in WebP, better portable graphics (BPG), and JPEG XL image compression algorithms. In8th Intl. Conf. on Comp. and Commun. Eng. (ICCCE). IEEE, 2021

    2021

  34. [42]

    Rafał Mantiuk, Dounia Hammou, and Param Hanji. HDR-VDP-3: A multi-metric for predicting image differences, quality and contrast distortions in high The JPEG XL Image Coding System: History, Features, Coding Tools, Design Rationale, and Future — 73/73 dynamic range and regular ...

    2023 arXiv

  35. [43]

    ColorVideoVDP: A visual difference predictor for image, video and dis- play distortions.ACM Trans

    Rafał Mantiuk, Param Hanji, Maliha Ashraf, Yuta Asano, and Alexandre Chapiro. ColorVideoVDP: A visual difference predictor for image, video and dis- play distortions.ACM Trans. Graph., 43(4), July 2024. [44]Rafał Mantiuk, Kil Joong Kim, Allan G Rempel, and Wolfgang Heidrich. H...

    2024

  36. [45]

    Marcellin, Michael J

    Michael W. Marcellin, Michael J. Gormish, Ali Bilgin, and Martin P. Boliek. An overview of JPEG-2000. InProceedings DCC 2000. Data compression confer- ence, pages 523–541. IEEE, 2000

    2000

  37. [46]

    Subjective image quality assessment with boosted triplet comparisons.IEEE Access, 9:138939– 138975, 2021

    Hui Men, Hanhe Lin, Mohsen Jenadeleh, and Diet- mar Saupe. Subjective image quality assessment with boosted triplet comparisons.IEEE Access, 9:138939– 138975, 2021

    2021

  38. [47]

    Addison-Wesley, 1999

    John Miano.Compressed image file formats: JPEG, PNG, GIF, XBM, BMP. Addison-Wesley, 1999

    1999

  39. [48]

    National Electrical Manufacturers Association, Rosslyn, V A, USA, 2024.http://www.dicomstandard.org/

    NEMA PS3 / ISO 12052.Digital Imaging and Com- munications in Medicine (DICOM) Standard. National Electrical Manufacturers Association, Rosslyn, V A, USA, 2024.http://www.dicomstandard.org/

    2024

  40. [49]

    Michael Niedermayer, Dave Rice, and J ´erˆome Martinez

    IETF RFC 9043. Michael Niedermayer, Dave Rice, and J ´erˆome Martinez. FFV1 video coding format versions 0, 1, and 3, 2021

    2021

  41. [50]

    Overview of TIFF 5.0

    Charles A Poynton. Overview of TIFF 5.0. InIm- age Processing and Interchange: Implementation and Systems, volume 1659, pages 152–158. SPIE, 1992

    1992

  42. [51]

    O’Reilly, 1999

    Greg Roelofs.PNG: the definitive guide. O’Reilly, 1999

    1999

  43. [52]

    Wiley, 2009

    Peter Schelkens, Athanassios Skodras, and Touradj Ebrahimi, editors.The JPEG 2000 suite. Wiley, 2009

    2000

  44. [53]

    Contemplating codec comparisons, De- cember 2022

    Jon Sneyers. Contemplating codec comparisons, De- cember 2022. https://cloudinary.com/blog/ contemplating-codec-comparisons

    2022

  45. [54]

    CID22: Large-scale subjective quality assessment for high fidelity image compression.TechRxiv, 2023

    Jon Sneyers, Elad Ben Baruch, and Yaron Vaxman. CID22: Large-scale subjective quality assessment for high fidelity image compression.TechRxiv, 2023

    2023

  46. [55]

    FLIF: Free lossless im- age format based on MANIAC compression

    Jon Sneyers and Pieter Wuille. FLIF: Free lossless im- age format based on MANIAC compression. In2016 IEEE international conference on image processing (ICIP), pages 66–70. IEEE, 2016

    2016

  47. [56]

    Improved re- dundancy reduction for JPEG files

    Matthias Stirner and Gerhard Seelmann. Improved re- dundancy reduction for JPEG files. InProc. of Picture Coding Symposium (PCS 2007), pages 7–9, 2007

    2007

  48. [57]

    Andrew Stockman and Lindsay T. Sharpe. The spec- tral sensitivities of the middle- and long-wavelength- sensitive cones derived from measurements in ob- servers of known genotype.Vision Research, 40(13):1711–1737, 2000

    2000

  49. [58]

    High efficiency video coding (HEVC).Inte- grated circuit and systems, algorithms and architec- tures, 39:40, 2014

    Vivienne Sze, Madhukar Budagavi, and Gary J Sul- livan. High efficiency video coding (HEVC).Inte- grated circuit and systems, algorithms and architec- tures, 39:40, 2014

    2014

  50. [59]

    JPEG Trust: an international standard facilitating the assessment of trustworthiness of digital media assets

    Frederik Temmermans, Sabrina Caldwell, Deepayan Bhowmik, and Touradj Ebrahimi. JPEG Trust: an international standard facilitating the assessment of trustworthiness of digital media assets. InApplications of Digital Image Processing XLVII, volume 13137, page 131370B. Internatio...

    2024

  51. [60]

    Adopt- ing the JPEG universal metadata box format for media authenticity annotations

    Frederik Temmermans and Leonard Rosenthol. Adopt- ing the JPEG universal metadata box format for media authenticity annotations. InApplications of Digital Im- age Processing XLIV, volume 11842, pages 165–170. SPIE, 2021

    2021

  52. [61]

    Fine-grained subjec- tive visual quality assessment for high-fidelity com- pressed images

    Michela Testolina, Mohsen Jenadeleh, Shima Moham- madi, Shaolin Su, Joao Ascenso, Touradj Ebrahimi, Jon Sneyers, and Dietmar Saupe. Fine-grained subjec- tive visual quality assessment for high-fidelity com- pressed images. InData Compression Conference (DCC). IEEE, 2025

    2025

  53. [62]

    On the assessment of high-quality images: advances on the JPEG AIC-3 activity

    Michela Testolina, Evgeniy Upenik, and Touradj Ebrahimi. On the assessment of high-quality images: advances on the JPEG AIC-3 activity. InApplications of Digital Image Processing XLVI, volume 12674, pages 180–190. SPIE, 2023

    2023

  54. [63]

    Trudeau, Nathan E

    Luc N. Trudeau, Nathan E. Egge, and David Barr. Pre- dicting chroma from luma in A V1.arXiv:1711.03951, 2018

    2018 arXiv

  55. [64]

    White paper: The benefits of adding JPEG-XL to the ISO PDF standard and PDF-Raster, 2025.https://twain.org/

    TWAIN Working Group. White paper: The benefits of adding JPEG-XL to the ISO PDF standard and PDF-Raster, 2025.https://twain.org/

    2025

  56. [65]

    PhD thesis, Univer- sit`a di Pisa, 2021

    Luca Versari.Compression Techniques for Large Graphs: Theory and Practice. PhD thesis, Univer- sit`a di Pisa, 2021

    2021

  57. [66]

    Cascading style sheets level 2 revision 1 (CSS 2.1) specification.http: //www.w3.org/TR/2011/REC-CSS2-20110607

    W3C Recommendation 07 June 2011. Cascading style sheets level 2 revision 1 (CSS 2.1) specification.http: //www.w3.org/TR/2011/REC-CSS2-20110607

    2011

  58. [67]

    Gregory K. Wallace. The JPEG still picture compres- sion standard.Commun. ACM, 34(4):30–44, 1991

    1991

  59. [68]

    Bovik, H.R

    Zhou Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simon- celli. Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing, 13(4):600–612, 2004

    2004

  60. [69]

    The LOCO-I lossless image compression al- gorithm: Principles and standardization into JPEG-LS

    Marcelo J Weinberger, Gadiel Seroussi, and Guillermo Sapiro. The LOCO-I lossless image compression al- gorithm: Principles and standardization into JPEG-LS. IEEE Transactions on Image Processing, 9(8):1309– 1324, 2000

    2000

  61. [70]

    A universal algo- rithm for sequential data compression.IEEE Transac- tions on information theory, 23(3):337–343, 1977

    Jacob Ziv and Abraham Lempel. A universal algo- rithm for sequential data compression.IEEE Transac- tions on information theory, 23(3):337–343, 1977

    1977

  62. [71]

    Performance evaluation of JPEG standards, WebP and PNG in terms of com- pression ratio and time for lossless encoding

    Emir ¨Ozt¨urk and Altan Mesut. Performance evaluation of JPEG standards, WebP and PNG in terms of com- pression ratio and time for lossless encoding. In2021 6th International Conference on Computer Science and Engineering (UBMK), pages 15–20, 2021

    2021

  63. [2023]

    https://ds.jpeg.org/whitepapers/ jpeg-xl-whitepaper.pdf

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.