Comments on this document can be sent to the PNG specification maintainers at png-info@uunet.uu.net or at png-list@dworkin.wustl.edu.
Distribution of this memo is unlimited.
At present, the latest version of this document is available on the
World Wide Web from
ftp://swrinde.nde.swri.edu/pub/png-group/documents/.
Permission is granted to copy and distribute this document for any purpose and without charge, provided that the copyright notice and this notice are preserved, and that any substantive changes or deletions from the original are clearly marked.
http://www.w3.org/pub/WWW/TR/WD-png.
Whenever you use any of these unregistered chunks you should also include a tEXt chunk describing it, for example:
tEXtComment\0 This file contains a zsCL chunk written according to the format given in Version 0.960330 of the PNG Proposed Chunks document.
Chunk name Multiple Ordering
OK? constraints
alIG No Before IDAT
drNG/DrNG No Before IDAT
faLS No Before IDAT
fiNG No None
loGE/LoGE No Before IDAT
spLT Yes Before IDAT
xySC No Before IDAT
zsCL No Before IDAT
This chunk's contents are
left 4 bytes left side of picture center 4 bytes center of picture (horizontal) right 4 bytes right side of picture top 4 bytes top of picture middle 4 bytes middle of picture (vertical) baseline 4 bytes typographic baseline of picture (vertical) bottom 4 bytes bottom of pictureAll values are signed 32-bit integer values, measured in pixels downward from the top of the image or rightward from the left edge of the image.
Applications will not normally try to use all of the alignment values at once. A browser might use the "center" value when it wishes to center the image left-to-right on a page. A typographic system might use the "baseline" value to line up an image that contains a fancy capital letter with the baseline of the adjacent text, or the "middle" value to line up images containing mathematical equations. Typographic systems could also determine the typographic "height", "depth", "width", and reference point values:
font_height = baseline - top font_width = right - left font_depth = bottom - baseline font_ref_x = left font_ref_y = baselineIf the alIG chunk is not present, applications can assume the following values:
left = 0 center = width/2 right = width top = 0 middle = height/2 baseline = 3/4 * height bottom = heightOnly one alIG chunk is permitted in a PNG datastream. If present, it must appear prior to the first IDAT chunk.
The chunk can be used as a critical chunk, named DrNG, or as an ancillary chunk, named drNG. The syntax and function is exactly the same whichever name is used. The critical version should only be used if the image cannot be meaningfully displayed without performing DrNG scaling. Decoders not recognizing DrNG will not attempt to display the image at all. Encoders are strongly encouraged to scale the image data properly and to use the noncritical version, drNG, if at all possible.
This chunk's contents are
min_sample m bytes Raw sample value corresponding to
minimum intensity of the graylevel
or red channel, written as a text
floating-point value [link to
Floating-Point Values in extensions
document]
0 1 byte a zero byte to separate fields
max_sample n bytes Raw sample value corresponding to
maximum intensity for the graylevel
or red channel
(The following entries are omitted for color_types 0 and 4)
0 1 byte a zero byte to separate fields
min_green m bytes raw sample value corresponding to
minimum green intensity
0 1 byte a zero byte to separate fields
max_green m bytes raw sample value corresponding to
maximum green intensity
0 1 byte a zero byte to separate fields
min_blue m bytes raw sample value corresponding to
minimum blue intensity
0 1 byte a zero byte to separate fields
max_blue m bytes raw sample value corresponding to
maximum blue intensity
If this chunk is present, graylevel or color sample values are to be
scaled for display between minimum and maximum intensity by linear
interpolation. When the sample value falls outside the range
min_value..max_value, it is to be set to min_value or max_value as
appropriate. For each graylevel or color sample, the conversion is
ratio := (2^bit_depth - 1)/(max_value - min_value) result := (input_sample - min_value) * ratio output_sample := LIMIT (0, result, 2^bit_depth-1) in which the function LIMIT (low, x, high) is MAX (MIN (x, high), low).Note that min_value and max_value are permitted to be negative, positive, or zero. The only restriction is that they be different from each other.
Encoders should not use drNG/DrNG in lieu of reasonably scaling the samples. For example, if the sample values range from 0 (black) to 1000 (white), it would be an extremely bad idea to use the drNG chunk to request the decoder to scale 1000 up to 65535, because a PNG viewer that does not understand drNG/DrNG (and, since this is an extension chunk, most viewers will probably not understand drNG/DrNG) would simply display a very dark rectangle. Instead, multiply your samples by 64, and use drNG with max_sample=64000 to request the decoder to do the final adjustment.
If the tRNS chunk is present, its value is compared to the unscaled pixel value, prior to applying the drNG/DrNG scaling. The faLS, gAMA, cHRM, and alpha conversions, if present, are applied to the scaled sample values.
If present, this chunk must appear before the first IDAT chunk. Only one drNG/DrNG chunk is permitted in a PNG file.
This chunk contains from 1 to 2^bitdepth 8-byte false-color palette entries:
index: 2 bytes, range 0 .. (2^bitdepth) -1) red: 2 bytes, range 0 .. 65535 green: 2 bytes, range 0 .. 65535 blue: 2 bytes, range 0 .. 65535 etc.The number of entries is determined from the chunk length. This length not divisible by 8 is an error.
This chunk may appear for color type 0 or color type 4. If it appears for any other color type, it will be ignored.
The complete 2^bitdepth-entry false-color palette can be built from the chunk data. If the first entry (index value 0) is missing, it will be assumed to be {0, 0, 0} (black). If the last entry (index value 2^bitdepth - 1) is missing, it will be assumed to be {65535, 65535, 65535} (white). The red, green, blue samples for other missing entries are filled in by linearly interpolating between the samples that are present, independently for each of the three color components.
Once the complete false-color palette is established, it is used similarly to PLTE. The first entry in the completed false-color palette is referenced by the grayscale value 0, the second by grayscale value 1, etc.
If the tRNS chunk is present, its value is compared to the graylevel, not to the converted false-color of the pixels. If the bKGD chunk is present, background pixels will be displayed in the false-color corresponding to the grayscale value found in the bKGD chunk. The gAMA, cHRM, and alpha conversions, if present, are applied to the color samples in the converted false-color pixels.
Note that the gAMA and other values must be selected so that the grayscale image is correctly displayed when the faLS chunk is unrecognized or ignored.
If this chunk does appear, it must precede the first IDAT chunk. There may be only one faLS chunk in a PNG file.
fingerprint 4 bytesThe fingerprint is the ADLER-32 check value computed on the image data, promoted by left-bit-replication (see bit_depth scaling in PNG spec) to 16-bit RGBA (colortype 6). Scaling is done without regard to the sBIT chunk, if present. It is calculated without regard to any ancillary chunks. The value is computed on the data in the same order as the bytes would appear in a color type 6 datastream, except that no filter bytes are included in the calculation. The ADLER-32 check value is computed as described in [link to adler-32 code in zlib docs]
An image's fingerprint will not change if the image is changed from interlaced to noninterlaced, or compressed with a different method, or filtered differently, or if any ancillary chunks are added, modified, or removed. It will change if there is any lossy conversion of the pixel data or if the PLTE data is changed in an indexed-color image.
While the fiNG chunk does not have any ordering requirements, You should put it early in the file for quick access.
The chunk can be used as a critical chunk, named LoGE, or as an ancillary chunk, named loGE. The syntax and function is exactly the same whichever name is used. The critical version should only be used if the image cannot be meaningfully displayed without performing LoGE scaling. Decoders not recognizing LoGE will not attempt to display the image at all. Encoders are strongly encouraged to include a gAMA chunk that permits a meaningful if not completely accurate display, and use the noncritical version, loGE, if at all possible.
The loGE/LoGE chunk's contents are
P0 n0 bytes (First parameter, a real number
written as a text floating-point
value [link to Floating-Point
Values in extensions document]
Null separator: 1 byte
P1 n1 bytes (Second parameter)
Null separator: 1 byte
P2 n2 bytes (Third parameter)
There is no trailing zero for the final string.
The scaling algorithm is
normalized_sample_value := input_sample/(2^input_bitdepth-1) scaled_value := P0 + P1 * P2^normalized_sample_value output_sample := LIMIT (0, scaled_value, 2^output_bitdepth-1) in which the function LIMIT (low, x, high) is MAX (MIN (x, high), low).For color_types 0 and 4, the normalized sample value is the grayscale value of the pixel normalized to the range [0.0:1.0] by dividing it by the maximum value for the input bit depth, using floating point arithmetic.
For color_types 2, 3, and 6, the scaling algorithm is applied independently to each of the color samples similarly normalized to [0.0:1.0].
The alpha channel, if present, is not affected by the loGE/LoGE transformation. Alpha compositing is done in the normal manner, with the transformed pixels forming the foreground image [link to Decoders: Alpha Processing in core spec]
Pure logarithmic data can be expressed with
P0 = 0 P1 = min_val P2 = max_val/min_valIn which the range [min_val..max_val] includes the minimum and maximum values appearing in the source data. When the image is decoded using the loGE/LoGE data, the gamma calculation for the decoded samples should be done as though the file_gamma were 1.0, regardless of the contents of the gAMA chunk.
It is advisable to include a gAMA chunk with logarithmic data, in case a viewer does not use the loGE data to decode it. A value of gamma should be chosen that allows the image to be displayed as well as it can be with a viewer that supports gAMA but not loGE.
You can select a value of gamma (also called file_gamma) by eye as described in [link to Gamma Tutorial: General Gamma Handling in core spec], or you can calculate one as follows:
If the maximum value of the logarithmically decoded pixels can reasonably be displayed as white on the monitor, then specifying a value of gamma given by
gamma = LN( LN(0.2) / LN(max_val/min_val) + 1) / LN(0.2) (in which LN() is the natural logarithm function)causes the two values max_val and 0.2*max_val to be reproduced at correct brightness on screen - these can be thought of as white and mid-grey. Tones between white and mid-grey will be a bit too bright, ones darker than mid-grey will be increasingly too dark, but it's a reasonable approximation for viewable display. A max_val/min_val ratio of 64 gives a gamma of about 0.3, a max_val/min_val of 1000 gives a gamma of about 0.165, so a viewer assuming a default gamma of 0.45 is going to give a bright, washed-out image for any log-encoded image.
In cases where the maximum sample value is many times brighter than scene white (i.e. the image is encoded to retain specular highlight information), the formula above doesn't apply, but there's probably no way of getting such an image to display reasonably anyway with a viewer that doesn't understand loGE.
If present, the loGE/LoGE chunk must appear before the first IDAT chunk. Only one instance of the loGE/LoGE chunk is permitted in a PNG datastream.
This chunk's contents are a zero-byte-terminated text string that names the palette, followed by a series of palette entries, each a ten-byte series, containing five unsigned integers:
name: n bytes (ASCII text)
null byte 1 byte (terminator)
red: 2 bytes (0 = black, 65535 = red)
green: 2 bytes (0 = black, 65535 = green)
blue: 2 bytes (0 = black, 65535 = blue)
alpha: 2 bytes (0 = fully transparent, 65535 = fully opaque)
frequency: 2 bytes (relative frequency of occurrence)
...
There can be any number of entries; a decoder determines the number of entries from the remaining chunk length after the null-terminated "name" string. This length not divisible by ten is an error. Entries must appear in decreasing order of "frequency".
The "name" (e.g. "rgba512 8-8-4-2 color cube", "rgb256 winter scenery", "rgb242 6-6-6 color cube plus 26 gray levels", "50-color rgb palette for use with early versions of Mosaic") identifies the palette, which may permit applications or people to choose the appropriate one when more than one suggested palette appears in a PNG file. The "name" string must consist only of printable ASCII characters and may not have leading or trailing blanks, but may have single embedded blanks. There must be at least one and no more than 79 characters in the name. Names are case-sensitive.
The red, green, and blue values are not premultiplied by alpha, nor are they precomposited against any background. A decoder can build a palette by compositing those palette entries against any background color or set of background colors that it chooses. See [link to bKGD]
Each frequency entry is proportional to the fraction of pixels in the image that are closest to that palette entry, without regard to any compositing against a background palette. The exact scale factor is chosen by the encoder, but should be chosen so that the range of individual values reasonably fills the range 0 to 65535. It is acceptable to artificially inflate the "frequency" values for "important" colors such as those in a company logo or in the facial features of a portrait. Zero is a valid value for frequency.
Note that the palette uses 16 bits (2 bytes) per value regardless of the image bit depth specification. Decoders wishing to construct 8-bit palettes can accomplish this by scaling down the RGB entries to 8 bits. [link to bit_depth rescaling].
Note: Earlier versions of the PNG specification recommended that the PLTE [link to PLTE] and hIST chunks be used for this purpose. While this is still allowed, to maintain backward compatibility, the spLT chunk is preferable, particularly when transparent pixels are present. When both the PLTE and spLT chunks are present, the PLTE data should only be used for decoding the indexed-color (color type 3) pixels, and the spLT data should be used for constructing the display palette. If the hIST chunk is also present, decoders that process the spLT chunk should ignore it.
This chunk may appear for any color type. If this chunk does appear, it must precede the first IDAT chunk. There may be multiple spLT chunks, with different names.
The xySC chunk's contents are
Xunit xu bytes (Latin-1 symbol or description of
the width unit, eg. milliseconds,
degrees West Longitude, etc).
A zero-length string may be used if
the data is dimensionless.
Null separator: 1 byte
Xoffset xo bytes X offset, a real number written as a
text floating-point value [link to
Floating-Point Values in extensions
document] This is the physical value
of "x" corresponding to the left
edge of the image.
Null separator: 1 byte
Xscale xs bytes X scale, the x-distance corresponding
to the width of a pixel. Must be
non-zero.
Null separator: 1 byte
Yunit yu bytes Latin-1 symbol or description of the
height unit, eg. meters, degrees North
Latitude, etc). A zero-length string
may be used if the data is
dimensionless.
Null separator: 1 byte
Yoffset yo bytes Y offset, a real number written as a
text floating-point value. This is
the physical value of "y"
corresponding to the top edge of the
image.
Null separator: 1 byte
Yscale ys bytes Y scale, the y-distance corresponding
to the height of a pixel. Must be
non-zero.
There is no trailing zero for the final string.
The scaling algorithm for finding the physical x-value and y-value of the middle of a pixel is
physical_x_value := Xoffset + Xscale * (column_number + 0.5) physical_y_value := Yoffset + Yscale * (row_number + 0.5)If present, the xySC chunk must appear before the first IDAT chunk. Only one instance of the xySC chunk is permitted in a PNG stream.
The zsCL chunk's contents are
Equation type 1 byte (0 for linear scaling,
1 for base-e exponential scaling,
2 for arbitrary-base exponential
scaling)
N 1 byte (Number of parameters)
Unit u bytes (Latin-1 symbol or description of
the unit, eg. K, Population Density,
MPa, etc). A zero-length string may
be used if the data is dimensionless.
Null separator: 1 byte
P0 p0 bytes (First parameter, a real number
written as a text floating-point
value [link to Floating-Point Values
in extensions document]
Null separator: 1 byte
P1 p1 bytes (Second parameter)
Null separator: 1 byte (only if P2 appears)
P2 p2 bytes (Third parameter, optional)
There is no trailing zero for the final string.
The scaling algorithm is
if equation_type == 0 then scaled_value := P0 + P1 * normalized_sample_value else if equation_type == 1 then scaled_value := P0 + P1 * EXP(P2 * normalized_sample_value) else if equation_type == 2 then scaled_value := P0 + P1 * P2^normalized_sample_valuein which EXP(x) is the exponential function, e^x, in which "e" is the base of natural logarithms (approximately 2.718282). For color_types 0 and 4, the normalized sample value is the grayscale sample value of the pixel normalized to the range [0.0:1.0] by dividing it by the maximum value for the bit depth, using floating point arithmetic.
For color_types 2, 3, and 6, the scaling algorithm is applied independently to each of the color sample values similarly normalized to [0.0:1.0].
Pure logarithmic data can be expressed either with
Equation_type = 1 Equation_type = 2 N = 3 or with N = 3 P0 = 0 P0 = 0 P1 = min P1 = min P2 = LOGe(max/min) P2 = max/minThe zsCL data is not intended to be used by a decoder to affect the way the pixels are displayed. zsCL is simply a comment, and could be used, for example, to construct a reference color bar scale beside the image, or to extract the original physical values recorded in the file.. The drNG/DrNG or loGE/LoGE chunk should be used if the encoder wants the decoder to modify the sample values for display purposes.
If present, the zsCL chunk must appear before the first IDAT chunk. Only one instance of the zsCL chunk is permitted in a PNG stream.
http://www.w3.org/pub/WWW/TR/WD-png.
The same precautions taken when displaying tEXt data should be taken when displaying the text contained in the "unit" strings of the xySC and zsCL chunks and in the "name" string of the spLT chunk. Viewers should not display these strings directly without first checking for the presence of nonprintable characters, and for the <ESC> character in particular.
No known additional security hazards are posed by the chunks described here.