ISO/IEC 18181-1:2022

INTERNATIONAL ISO/IEC
STANDARD 18181-1
First edition
2022-03
Information technology — JPEG XL
image coding system —
Part 1:
Core coding system
Technologies de l'information — Systѐme de codage d'images
JPEG XL —
Partie 1: Système de codage de noyau
Reference number
© ISO/IEC 2022
© ISO/IEC 2022
All rights reserved. Unless otherwise specified, or required in the context of its implementation, no part of this publication may
be reproduced or utilized otherwise in any form or by any means, electronic or mechanical, including photocopying, or posting on
the internet or an intranet, without prior written permission. Permission can be requested from either ISO at the address below
or ISO’s member body in the country of the requester.
ISO copyright office
CP 401 • Ch. de Blandonnet 8
CH-1214 Vernier, Geneva
Phone: +41 22 749 01 11
Email: copyright@iso.org
Website: www.iso.org
Published in Switzerland
ii
© ISO/IEC 2022 – All rights reserved

Contents Page
Foreword .v
Introduction . vi
1 Scope . 1
2 Normative references . 1
3 Terms and definitions . 1
3.1 Data storage . 2
3.2 Inputs . 2
3.3 Processes . 3
3.4 Image organization . 4
3.5 DCT . 5
4 Abbreviated terms . 6
5 Conventions . 6
5.1 Mathematical symbols . 6
5.2 Functions . 6
5.3 Operators . 7
5.4 Pseudocode . 7
6 Functional concepts . .8
6.1 Image organization . 8
6.2 Group splitting . 8
6.3 Codestream and bitstream . 9
6.4 Multiple frames . 10
6.5 Mirroring . 10
7 Encoder requirements .10
8 Decoder requirements.10
9 Codestream .10
9.1 Syntax . 10
9.1.1 Reading a field . 11
9.1.2 Initializing a field . 11
9.2 Field types . 11
9.2.1 u(n) . 11
9.2.2 U32(d0, d1, d2, d3) . 11
9.2.3 U64() . 11
9.2.4 Varint() . 12
9.2.5 U8() . 12
9.2.6 F16() . 12
9.2.7 Bool() .12
9.2.8 Enum(EnumTable) .12
9.2.9 ZeroPadToByte() . 13
9.3 Structure . 13
10 Decoding process .13
Annex A (normative) Headers .15
Annex B (normative) ICC profile . .25
Annex C (normative) Frames .32
Annex D (normative) Entropy decoding .58
Annex E (normative) Weighted predictor .67
Annex F (normative) Adaptive quantization . .70
Annex G (normative) Chroma from luma.71
iii
© ISO/IEC 2022 – All rights reserved

Annex H (normative) Extensions .72
Annex I (normative) Integral transforms .73
Annex J (normative) Restoration filters . .84
Annex K (normative) Image features .87
Annex L (normative) Colour transforms .92
Annex M (informative) Encoder overview .98
Bibliography . 101
iv
© ISO/IEC 2022 – All rights reserved

Foreword
ISO (the International Organization for Standardization) and IEC (the International Electrotechnical
Commission) form the specialized system for worldwide standardization. National bodies that are
members of ISO or IEC participate in the development of International Standards through technical
committees established by the respective organization to deal with particular fields of technical
activity. ISO and IEC technical committees collaborate in fields of mutual interest. Other international
organizations, governmental and non-governmental, in liaison with ISO and IEC, also take part in the
work.
The procedures used to develop this document and those intended for its further maintenance
are described in the ISO/IEC Directives, Part 1. In particular, the different approval criteria
needed for the different types of document should be noted. This document was drafted in
accordance with the editorial rules of the ISO/IEC Directives, Part 2 (see www.iso.org/directives or
www.iec.ch/members_experts/refdocs).
Attention is drawn to the possibility that some of the elements of this document may be the subject
of patent rights. ISO and IEC shall not be held responsible for identifying any or all such patent
rights. Details of any patent rights identified during the development of the document will be in the
Introduction and/or on the ISO list of patent declarations received (see www.iso.org/patents) or the IEC
list of patent declarations received (see https://patents.iec.ch).
Any trade name used in this document is information given for the convenience of users and does not
constitute an endorsement.
For an explanation of the voluntary nature of standards, the meaning of ISO specific terms and
expressions related to conformity assessment, as well as information about ISO's adherence to
the World Trade Organization (WTO) principles in the Technical Barriers to Trade (TBT) see
www.iso.org/iso/foreword.html. In the IEC, see www.iec.ch/understanding-standards.
This document was prepared by Joint Technical Committee ISO/IEC JTC 1, Information technology,
Subcommittee SC 29, Coding of audio, picture, multimedia and hypermedia information.
A list of all parts in the ISO/IEC 18181 series can be found on the ISO and IEC websites.
Any feedback or questions on this document should be directed to the user’s national standards
body. A complete listing of these bodies can be found at www.iso.org/members.html and
www.iec.ch/national-committees.
v
© ISO/IEC 2022 – All rights reserved

Introduction
The International Organization for Standardization (ISO) and International Electrotechnical
Commission (IEC) draw attention to the fact that it is claimed that compliance with this document may
involve the use of a patent.
ISO and IEC take no position concerning the evidence, validity and scope of this patent right.
The holder of this patent right has assured ISO and IEC that they are willing to negotiate licences under
reasonable and non-discriminatory terms and conditions with applicants throughout the world. In this
respect, the statement of the holder of this patent right is registered with ISO and IEC. Information may
be obtained from the patent database available at www.iso.org/patents.
Attention is drawn to the possibility that some of the elements of this document may be the subject
of patent rights other than those in the patent database. ISO and IEC shall not be held responsible for
identifying any or all such patent rights.
vi
© ISO/IEC 2022 – All rights reserved

INTERNATIONAL STANDARD ISO/IEC 18181-1:2022(E)
Information technology — JPEG XL image coding system —
Part 1:
Core coding system
1 Scope
This document defines a set of compression methods for coding one or more images of bi-level,
continuous-tone greyscale, or continuous-tone colour, or multichannel digital samples.
This document:
— specifies decoding processes for converting compressed image data to reconstructed image data;
— specifies a codestream syntax containing information for interpreting the compressed image data;
— provides guidance on encoding processes for converting source image data to compressed image
data.
2 Normative references
The following documents are referred to in the text in such a way that some or all of their content
constitutes requirements of this document. For dated references, only the edition cited applies. For
undated references, the latest edition of the referenced document (including any amendments) applies.
ISO 15076-1:2010, Image technology colour management — Architecture, profile format and data
structure — Part 1: Based on ICC.1:2010
ISO/IEC 60559, Information technology — Microprocessor Systems — Floating-Point arithmetic
IEC 61966-2-1, Multimedia systems and equipment — Colour measurement and management — Part 2-1:
Colour management — Default RGB colour space — sRGB
Rec. ITU-R BT.2100-2, Image parameter values for high dynamic range television for use in production and
international programme exchange
Rec. ITU-R BT.709-6, Parameter values for the HDTV standards for production and international
programme exchange
SMPTE ST 428-1, D-Cinema distribution master — Image characteristics
3 Terms and definitions
For the purposes of this document, the following terms and definitions apply.
ISO and IEC maintain terminology databases for use in standardization at the following addresses:
— ISO Online browsing platform: available at https:// www .iso .org/ obp
— IEC Electropedia: available at https:// www .electropedia .org/
© ISO/IEC 2022 – All rights reserved

3.1 Data storage
3.1.1
byte
8 consecutive bits encoding a value between 0 and 255
3.1.2
big endian
value representation with bytes in most to least-significant order
3.1.3
bitstream
sequence of bytes from which bits are read starting from the least-significant bit of the first byte
3.1.4
codestream
bitstream representing compressed image data
3.1.5
bundle
structured data consisting of one or more fields
3.1.6
field
numerical value or bundle, or an array of either
3.1.7
histogram
array of unsigned integers representing a probability distribution, used for entropy coding
3.1.8
set
unordered collection of elements
3.2 Inputs
3.2.1
pixel
vector of dimension corresponding to the number of channels, consisting of samples
3.2.2
sample
integer or real value, of which there is one per channel per pixel
3.2.3
greyscale
image representation in which each pixel is defined by a single sample representing intensity (either
luminance or luma depending on the ICC profile)
3.2.4
continuous-tone image
image having samples consisting of more than one bit
3.2.5
opsin
photosensitive pigments in the human retina, having dynamics approximated by the XYB colour space
3.2.6
burst
sequences of images typically captured with identical settings
© ISO/IEC 2022 – All rights reserved

3.2.7
animation
series of pictures and timing delays to display as a video medium
3.2.8
composite
series of images that are superimposed
3.2.9
frame
single image (possibly part of a burst or animation or composite)
3.2.10
preview
lower-fidelity rendition of one of the frames (e.g. lower resolution), or a frame that represents the entire
content of all frames
3.3 Processes
3.3.1
decoding process
process which takes as its input a codestream and outputs a continuous-tone image
3.3.2
decoder
embodiment of a decoding process
3.3.3
encoding process
process which takes as its input continuous-tone image(s) and outputs compressed image data in the
form of a codestream
3.3.4
encoder
embodiment of an encoding process
3.3.5
lossless
descriptive term for encoding and decoding processes in which the output of a decoding procedure is
identical to the input to the encoding procedure
3.3.6
lossy
descriptive term for encoding and decoding processes which are not lossless
3.3.7
upsampling
procedure by which the (nominal) spatial resolution of a channel is increased
3.3.8
downsampling
procedure by which the spatial resolution of a channel is reduced
3.3.9
entropy encoding
lossless procedure designed to convert a sequence of input symbols into a sequence of bits such that the
average number of bits per symbol approaches the entropy of the input symbols
3.3.10
entropy encoder
embodiment of an entropy encoding procedure
© ISO/IEC 2022 – All rights reserved

3.3.11
entropy decoding
lossless procedure which recovers the sequence of symbols from the sequence of bits produced by the
entropy encoder
3.3.12
entropy decoder
embodiment of an entropy decoding procedure
3.3.13
Gabor-like transform
convolution with default or signalled 3x3 kernel for deblocking
3.3.14
tick
unit of time such that animation frame durations are integer multiples of the tick duration
3.4 Image organization
3.4.1
grid
2-dimensional array; a [x, y] means addressing an element of grid a at row y and column x. Where so
specified, addressing elements with coordinates outside of bounding rectangle (x < 0, or y < 0, or x >=
width, or y >= height) is allowed
3.4.2
sample grid
common coordinate system for all samples of an image, with top-left coordinates (0, 0), the first
coordinate increasing towards the right, and the second increasing towards the bottom
3.4.3
channel
component
rectangular array of samples having the same designation, regularly aligned along a sample grid
3.4.4
rectangle
rectangular area within a channel or grid
3.4.5
width
width in samples of a sample grid or a rectangle
3.4.6
height
height in samples of a sample grid or a rectangle
3.4.7
raster order
access pattern from left to right in the top row, then in the row below and so on
3.4.8
naturally aligned
positioning of a power-of-two sized rectangle such that its top and left coordinates are divisible by its
width and height, respectively
3.4.9
block
naturally aligned square rectangle covering up to 8 × 8 input pixels
© ISO/IEC 2022 – All rights reserved

3.4.10
group
n n
naturally aligned square rectangle covering up to 2 × 2 (with n between 7 and 10, inclusive) input
pixels
3.4.11
table of contents
data structure that enables seeking to a group or the next frame within a codestream
3.4.12
section
part of the codestream with an offset and length that are stored in a frame's table of contents
3.5 DCT
3.5.1
coefficient
input value to the inverse DCT
3.5.2
quantization
method of reducing the precision of individual coefficients
3.5.3
varblock
variable-size rectangle of input pixels
3.5.4
dct_block
an array with 64 elements corresponding to DCT coefficients of a (8 × 8) block
3.5.5
var-DCT
lossy encoding of a frame that applies DCT to varblocks
3.5.6
LF coefficient
lowest frequency DCT coefficient, containing the average value of a block or the lowest-frequency
coefficient within the 8 × 8 rectangle of a varblock of size greater than 8 × 8
3.5.7
HF coefficients
all DCT coefficients apart from the LF coefficients, i.e. the high frequency coefficients
3.5.8
pass
data enabling decoding of successively higher resolutions
3.5.9
LF group
n n n+3 n+3
2 × 2 LF values from a naturally aligned rectangle covering up to 2 × 2 input pixels
3.5.10
quantization weight
factor that a quantized coefficient is multiplied by prior to application of the inverse DCT in the decoding
process
3.5.11
channel decorrelation
method of reducing total encoded entropy by removing correlations between channels
© ISO/IEC 2022 – All rights reserved

3.5.12
channel correlation factor
factor by which a channel should be multiplied by before adding it to another channel to undo the
channel decorrelation process
4 Abbreviated terms
DCT: discrete cosine transform (DCT-II as specified in I.2)
IDCT: inverse discrete cosine transform (DCT-III as specified in I.2)
LF: N / 8 × M / 8 square of lowest frequency coefficients of N × M DCT coefficients
RGB: additive colour model with red, green, blue channels
LMS: absolute colour space representing the response of cone cells in the human eye
XYB: absolute colour space based on gamma-corrected LMS, in which X is derived from the difference
between L and M, Y is an average of L and M (behaves similarly to luminance), and B is derived from the
S ("blue") channel
5 Conventions
5.1 Mathematical symbols
[a, b], (c, d), [e, f)
closed or open or half-open intervals containing all integers or real
numbers x (depending on context) such that
a ≤ x ≤ b, c < x < d, e ≤ x < f.
{a, b, c}
ordered sequence of elements
π
the smallest positive zero of the sine function
5.2 Functions
sqrt(x)
square root, such that (sqrt(x)) == x and sqrt(x) >= 0. Undefined for x < 0.
cbrt(x) 3
cube root, such that (cbrt(x)) == x.
cos(r)
cosine of the angle r (in radians)
erf(x)
x
−t
Gauss error function: erf(x) = edt
∫
π
log(x)
natural logarithm of x. Undefined for x <= 0.
log2(x)
base-two logarithm of x. Undefined for x <= 0.
floor(x)
the largest integer that is less than or equal to x
ceil(x)
the smallest integer that is greater than or equal to x
abs(x)
absolute value of x: equal to -x if x < 0, otherwise x
sign(x)
sign of x, 0 if x is 0, +1 if x is positive, -1 if x is negative
UnpackSigned(u)
equivalent to u / 2 if u is even, and -(u + 1) / 2 if u is odd
© ISO/IEC 2022 – All rights reserved

clamp(x, lo, hi)
equivalent to min({max({lo, x}), hi})
InterpretAsF16(u)
the real number resulting from interpreting the unsigned 16-bit integer u as a
binary16 floating-point number representation (cf. ISO/IEC 60559)
InterpretAsF32(u)
the real number resulting from interpreting the unsigned 32-bit integer u as a
binary32 floating-point number representation (cf. ISO/IEC 60559)
len(a)
length (number of elements) of array a
sum(a)
sum of all elements of the array/tuple/sequence a
max(a)
maximal element of the array/tuple/sequence a
min(a)
smallest element of the array/tuple/sequence a
5.3 Operators
[2]
This document uses the operators defined by the C++ programming language , with the following
differences:
×
multiplication
×=
a ×= b is equivalent to a = a × b
/
division of real numbers without truncation or rounding. Division by zero is undefined.
y
x
exponentiation, x to the power of y
<<
s
left shift: x << s is defined as x × 2
>> s
right shift: x >> s is defined as floor(x / 2 )
Umod
a Umod d is the unique integer r in [0, d) for which a == r + q×d for a suitable integer q
Idiv
a Idiv b is equivalent to a / b, rounded towards zero to an integer value
The order of precedence for these operators is listed below in descending order. If several operators
appear in the same line, they have equal precedence. When several operators of equal precedence
appear at the same level in an expression, evaluation proceeds according to the associativity of the
operator (either from right to left or from left to right).
Operators Type of operation Associativity
++x, --x
prefix increment/decrement right to left
y
x
exponentiation right to left
!, ~
logical/bitwise NOT right to left
×, /, Idiv, Umod
multiplication, division, integer division, remainder left to right
+, -
addition and subtraction left to right
<<, >>
left shift and right shift left to right
< , >, <=, >=
relational left to right
=
assignment right to left
+=, -=, ×=
compound assignment right to left
5.4 Pseudocode
This document describes functionality using pseudocode formatted as follows:
© ISO/IEC 2022 – All rights reserved

// Informative comment
var = u(8); // Defined in 9.2.1
if (var == 1) return; // Stop executing this code snippet
[[Normative specification: var != 0]]
(out1, out2) = Function(var, kConstant);
Variables such as var are typically referenced by text outside the source code.
[2]
The semantics of this pseudocode are those of the C++ programming language , with the following
exceptions:
— Symbols from 5.1 and functions from 5.2 are allowed;
— Multiplication, division, remainder and exponentiation are expressed as specified in 5.3;
— Functions can return tuples which unpack to variables as in the above example;
— [[ ]] enclose normative directives specified using prose;
— All integers are stored using two's complement;
— Expressions and variables of which types are omitted, are understood as real numbers.
Where unsigned integer wraparound and truncated division are required, Umod and Idiv (see 5.3) are
used for those purposes.
Numbers with a 0x prefix are in base 16 (hexadecimal), and apostrophe (') characters inside them are
understood to have no effect.
EXAMPLE 0x0001'0000 == 65536.
6 Functional concepts
6.1 Image organization
A channel is defined as a rectangular array of (integer or real) samples regularly aligned along a sample
grid of width sample positions horizontally and height sample positions vertically. The number of
channels may be 1 to 4099 (see num_extra_channels in A.6).
A pixel is defined as a vector of dimension corresponding to the number of channels, consisting of
samples with a position matching that of the pixel. The index of a sample is numbered from 0 to number
of channels - 1.
An image is defined as the two-dimensional array of pixels, and its width is width and height is height.
Unless otherwise mentioned, channels are accessed in the following "raster order": left to right column
within the topmost row, then left to right column within the row below the top, and so on until the
rightmost column of the bottom row.
6.2 Group splitting
Channels are logically partitioned into naturally-aligned groups of kGroupDim × kGroupDim samples.
The effective dimension of a group (i.e. how many pixels to read) can be smaller than kGroupDim for
groups on the right or bottom of the image. The decoder ensures the decoded image has the dimensions
specified in SizeHeader by cropping at the right and bottom as necessary. Unless otherwise specified,
kGroupDim is 256.
LF groups likewise consist of kGroupDim × kGroupDim LF samples, with the possibility of a smaller
effective size on the right and bottom of the image.
© ISO/IEC 2022 – All rights reserved

Groups can be decoded independently. A 'table of contents' stores the size (in bytes) of each group to
allow seeking to any group. An optional permutation allows groups to be arranged in arbitrary order
within the codestream.
EXAMPLE Figure 1 shows an example of the HF groups and LF groups of an image.

Frame: 2970×1868 pixels
HF groups:
11×7 groups of 256×256 pixels,
1×7 groups of 154×256 pixels,
7×1 groups of 256×76 pixels,
1 group of 154×76 pixels
LF groups:
1 group of 256×233 LF coefficients
(covering 2048×1868 pixels),
1 group of 116×233 LF coefficients
(covering 922×1868 pixels)
Figure 1 — Group splitting example
6.3 Codestream and bitstream
A bitstream is a finite sequence of bytes. A codestream is a bitstream that represent compressed image
data and metadata. N bytes can also be viewed as 8 × N bits. The first 8 bits are the bits constituting the
first byte, in least to most significant order, the next eight bits (again in least to most significant order)
constitute the second byte, and so on. Unless otherwise specified, bits are read from the codestream as
specified in 9.2.1.
NOTE Ordering bits from least to most significant allows using special CPU instructions to isolate the least-
significant bits.
Subsequent Annexes or subclauses indicate some elements of the codestream are byte-aligned. For
such elements, the decoder takes actions before and after reading the element as follows. Immediately
before encountering the element, the decoder invokes ZeroPadToByte() (9.2.9). After finishing reading
the element, the decoder invokes ZeroPadToByte() (9.2.9).
© ISO/IEC 2022 – All rights reserved

ZeroPadToByte specifies that the padding bits, if any, are zero for the codestream to be valid. This can
serve as an additional indicator of codestream integrity.
6.4 Multiple frames
A codestream may contain multiple frames. These can constitute an animation, a burst (arbitrary
images with identical dimensions), or a composite still image with one or more frames rendered on top
of the first frame.
NOTE The frame that is being decoded is referred to as the current frame.
6.5 Mirroring
Some operations access samples with coordinates cx, cy that are outside the image bounds. The decoder
redirects such accesses to a valid sample at the coordinates Mirror1D(cx, width), Mirror1D(cy,
height), defined in the following code:
Mirror1D(coord, size) {
if (coord < 0) return Mirror1D(-coord - 1, size);
else if (coord >= size) return Mirror1D(2 × size - 1 - coord, size);
else return coord;
}
7 Encoder requirements
An encoder is an embodiment of the encoding process. This document does not specify an encoding
process, and any encoding process is acceptable as long as the codestream conforms to the codestream
syntax specified in this document. Annex M provides an informative description of such an encoding
process.
8 Decoder requirements
A decoder is an embodiment of the decoding process. The decoder reconstructs sample values
(arranged on a rectangular sampling grid) from a codestream as specified in this document. Annexes A
to L are normative in the sense that they are defining an output that alternative implementations shall
duplicate.
9 Codestream
9.1 Syntax
The codestream is organized into "bundles" consisting of one or more conceptually related "fields". A
field can be a bundle, or an array/value of a type from 9.2. The graph of contained-within relationships
between types of bundles is acyclic. This document specifies the structure of each bundle using tables
structured as in Table 1, with one row per field (in top to bottom order).
Table 1 — Structure of a table describing a bundle
condition type default name
If condition is blank or evaluates to true, the field is read from the codestream (9.1.1). Otherwise, the
field is instead initialized to default (9.1.2).
A condition of the form for(i = 0; condition; ++i) is equivalent to replacing this row with a
sequence of rows obtained from the current one by removing its condition and replacing the value of i
© ISO/IEC 2022 – All rights reserved

in each column with the consecutive values assumed by the loop-variable i until condition is false. If
condition is initially false, the current row has no effect.
If name ends with [n] then the field is a fixed-length array with n entries. If condition is blank or
evaluates to true, each array element is read from the codestream (9.1.1) in order of increasing index.
Otherwise, each element is initialized to default (9.1.2). The name is potentially referenced in the
condition or type of a subsequent row, or the condition of the same row.
9.1.1 Reading a field
If a field is to be read from the codestream, the type entry determines how to do so. If it is a basic field
type, it is read as described in 9.2. Otherwise type is a (nested) bundle denoted Nested, residing within
a parent bundle Parent. Nested is read as if the rows of the table defining Nested were inserted into
the table defining Parent in place of the row that defined Nested. This principle is applied recursively,
corresponding to a depth-first traversal of all fields.
9.1.2 Initializing a field
If a field is to be initialized to default, the type entry determines how to do so. If it is a bundle, then
default is blank and each field of the bundle is (recursively) initialized to the default specified within
the bundle's table. Otherwise, if default is not blank, the field is set to default, which is a valid value of
the same type.
9.2 Field types
9.2.1 u(n)
u(0) evaluates to the value zero without reading any bits. For n > 0, u(n) reads n bits from the
n
codestream, advances the position accordingly, and returns the value in the range [0, 2 ) represented
by the bits. The decoder first reads the least-significant bit of the value, from the least-significant not
yet consumed bit in the first not yet fully consumed byte of the codestream. The next bit of the value
(in increasing order of significance) is read from the next (in increasing order of significance) bit of the
same byte unless all its bits have already been consumed, in which case the decoder reads from the
least-significant bit of the next byte, and so on.
9.2.2 U32(d0, d1, d2, d3)
In this subclause, 'distribution' refers to one of the following three encodings of a range of values: Val(u),
Bits(n), or BitsOffset(n, offset). The d0, d1, d2, d3 parameters represent distributions.
U32(d0, d1, d2, d3) reads an unsigned 32-bit value in [0, 2 ) as follows. The decoder first reads a u(2)
from the codestream indicating which distribution to use (0 selects d0, 2 selects d2 etc.). Let d denote
this distribution, which determines how to decode value.
If d is Val(u), value is the integer u. If d is Bits(n), value is read as u(n). If d is BitsOffset(n, offset), the
decoder reads v = u(n). The resulting value is (offset + v) Umod 2 .
NOTE The value of u is implicitly defined by Val(u) and not stored in the codestream.
EXAMPLE For a field of type U32(Val(8), Val(16), Val(32), Bits(7)), the bits 10 result in value = 32. For a
U32(Bits(2), Bits(4), Bits(6), Bits(8)) field, the bits 010111 result in value = 7.
9.2.3 U64()
U64() reads an unsigned 64-bit value in [0, 2 ) using a single variable-length encoding. The decoder
first reads a u(2) selector s. If s == 0, value is 0. If s == 1, value is BitsOffset(4, 1) (9.2.2). If s == 2, value
is BitsOffset(8, 17). Otherwise s == 3 and value is read from a 12-bit part, zero or more 8-bit parts, and
zero or one 4-bit part as specified by the following code:
© ISO/IEC 2022 – All rights reserved

value = u(12); shift = 12;
while (u(1) == 1) {
if (shift == 60) {
value += u(4) << shift; // only 4, we already read 60
break;
}
value += u(8) << shift; shift += 8;
}
EXAMPLE the largest possible value (2 - 1) is encoded as 73 consecutive 1-bits.
9.2.4 Varint()
Varint() reads an unsigned integer value of up to 63 bits as specified by the following code:
value = 0; shift = 0;
while (1) {
b = u(8);
value += (b & 127) << shift;
if (b <= 127) break;
shift += 7;  [[ shift < 63 ]];
}
9.2.5 U8()
U8() reads an integer value in the range [0, 256) using [1, 12) bits as specified by the following code:
if (u(1) == 0) value = 0;
else { n = u(3); value = u(n) + (1 << n); }
EXAMPLE The bit 0 results in value 0, bits 1000 result in value 1, bits 10011 in value 3.
9.2.6 F16()
F16() reads a binary16 representation (as specified in ISO/IEC 60559) of a real value in [-65504, 65504].
value is as specified by the following code:
bits16 = u(16);
biased_exp = (bits16 >> 10) & 0x1F;
value = InterpretAsF16(bits16);
The value of biased_exp is not 31.
NOTE This rules out NaN and infinities.
9.2.7 Bool()
Bool() reads a boolean value as u(1) ? true : false.
9.2.8 Enum(EnumTable)
Enum(EnumTable) reads v = U32(Val(0), Val(1), BitsOffset(4, 2), BitsOffset(6, 18)). The value v does
not exceed 63 and is a value defined by the table in the subclause titled EnumTable. Such tables are
structured according to Table 2, with one row per unique value.
© ISO/IEC 2022 – All rights reserved

Table 2 — Structure of a table describing an enumerated type
name value meaning
An enumerated type is interpreted as having the meaning of the row where value is v. name (or where
ambiguous, EnumTable.name) is an identifier for purposes of referring to the same meaning elsewhere in
this document. name begins with a k prefix, e.g. kRGB.
9.2.9 ZeroPadToByte()
The decoder reads a u(n), where n is zero if P, the 0-based index of the next unread bit in the codestream,
is a multiple of 8, otherwise n = 8 - (P Umod 8). The result is equal to zero.
NOTE The effect of ZeroPadToByte() is skip to the next byte boundary if not already at a byte boundary, and
require all skipped bits to have value 0.
9.3 Structure
The codestream consists of headers (Annexes A, H) and an ICC profile, if present (Annex B), followed by
one or more frames (Annex C), as shown in Table 3. This table and those it references, together with the
syntax description (9.1), specify how a decoder reads the headers and frame(s).
Table 3 — Codestream structure
condition type name Annex
headers
Headers A, H
headers.metadata.want_icc icc
B
headers.metadata.have_preview preview_frame
Frame C
Frame frames[0] C
for (i = 1; !frames[i - 1].frame_header.is_last; ++i)
Frame frames[i] C
The preview_frame is a self-contained frame whose FrameHeader (C.2) specifies lf_level=0, frame_
type=kRegularFrame and save_as_reference=0.
10 Decoding process
After reading the header and frame data (9.3 and Annexes A, B and C), the decoder can be viewed as a
pipeline with multiple stages, as shown in Figure 2.
Some Annexes or subclauses begin with a condition; the decoder only applies the processing steps
described in such an Annex or subclause if its condition holds true.
Annex D specifies how the decoder reads entropy-coded data.
Annex E specifies how the decoder transforms channels decoded as residuals of a self-correcting
weighted predictor into reconstructed samples.
The decoder converts adaptively quantized integers to DCT coefficients as specified in Annex F.
Annex G specifies how the decoder recorrelates channels stored as differences w.r.t. a linear function of
another channel for the purpose of decorrelation ('chroma from luma').
The decoder performs or skips inverse integral transforms as specified in Annex I.
The decoder applies zero, one or two or restoration filters as specified in Annex J.
The presence/absence of additional image features (patches, splines and noise) is indicated in the frame
header. The decoder draws these as specified in Annex K. Image features (if present) are rendered after
restoration filters (if enabled), in the listed order.
© ISO/IEC 2022 – All rights reserved

Finally, the decoder performs zero, one or more colour-space transforms as specified in Annex L.
NOTE For an introduction to the coding tools and additional background, refer to the Joint Photographic
[5]
Experts Group (JPEG) committee JPEG XL website .
EXAMPLE The lossless mode does not involve adaptive dequantization.
Key
DQ dequantization
CfL chroma from luma
IT integral transform
RF restoration filter
PT patches
SP splines
NS noise
CT colour transform
Figure 2 — Decoder block diagram after LF/HF coefficients of a frame have been read
© ISO/IEC 2022 – All rights reserved

Annex A
(normative)
Headers
A.1 General
The decoder reads the Headers bundle (Table A.1) as specified in 9.1.
Table A.1 — Headers bundle
condition type name subclause
signature
Signature A.2
size
SizeHeader A.3
metadata
ImageMetadata A.6
A.2 Signature
Table A.2 specifies the Signature bundle.
Table A.2 — Signature bundle
condition type default name
ff
u(8)
type
u(8)
ff is 255. type is 10.
A.3 Image dimensions
Table A.3 specifies the SizeHeader bundle.
Table A.3 — SizeHeader bundle
condition type default name
small
Bool() false
small height_div8_minus_1
u(5) 0
height_minus_1
!small U32(Bits(9), Bits(13), Bits(18), Bits(30)) 0
ratio
u(3) 0
width_div8_minus_1
small && ratio == 0 u(5) 0
width_minus_1
!small && ratio == 0 U32(Bits(9), Bits(13), Bits(18), Bits(30)) 0
height is defined as small ? (height_div8_
...