[
https://issues.apache.org/jira/browse/FOP-2847?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tobias Hain updated FOP-2847:
-----------------------------
Description:
h4. TL;NR
When using palette-based transparency via tRNS chunk in PNG images, the
generated PDFs are broken.
h4. palette-based with transparency.
{quote}*8.5.2. Palette-Based with Transparency*
"The PNG spec forbids the use of a full alpha channel with palette-based
images, but it does allow `cheap alpha` via the transparency chunk, tRNS [...]
In effect, it transforms the palette from an RGB lookup table to an RGBA table
[...] a single PNG transparency entry should come first"
{quote}
[http://www.libpng.org/pub/png/book/chapter08.html#png.ch08.div.5.2]
h4. general PDF encoding strategy
Option 1:
{quote}most libraries just undo the palettization and use a full color image +
mask.
{quote}
[https://forums.adobe.com/message/6148131#6148131]
Option 2:
{quote}SMask would be simpler since you could leave the palette in place
{quote}
[https://forums.adobe.com/message/6148557#6148557]
h4. what does FOP do
It creates a chroma-key mask:
{code:java}
/Type /XObject
/ColorSpace [/Indexed /DeviceRGB 255 <D9DB ...>]
/Mask [182 182 151 151 158 158]
{code}
Acrobat Reader even fails to read those files while PDF viewers embedded into
Chrome/Firefox read the file, but show visual glitches.
FOP doesn't implement Option 1 yet. None of the ImageLoaders is prepared to do
that:
[https://wiki.apache.org/xmlgraphics-fop/HowTo/ImageLoaderRawPNG]
In fact the following results in missing Option 2:
{code:java}
public void setup(PDFDocument doc) {
RenderedImage ri = getImage().getRenderedImage();
// code deleted
Raster raster = GraphicsUtil.getAlphaRaster(ri);
if (raster != null) {
AlphaRasterImage alphaImage = new AlphaRasterImage("Mask:" + getKey(),
raster);
this.softMask = doc.addImage(null, alphaImage).makeReference();
}
{code}
[https://github.com/apache/fop/blob/ff452f3685a02715aee791d651b58dd1797ddfd4/fop-core/src/main/java/org/apache/fop/render/pdf/ImageRenderedAdapter.java]
In the above code raster will be null and therefore FOP fails to initialize a
softMask. {{GraphicsUtil.getAlphaRaster()}} calls internally
{{ColorModel.getAlphaRaster()}}, which returns null:
{quote}If this is an IndexColorModel which has alpha in the lookup table, this
method will return null since there is no spatially discrete alpha channel.
{quote}
[https://docs.oracle.com/javase/8/docs/api/index.html?java/awt/image/ColorModel.html]
This code will then add the chroma key mask to the image:
{code:java}
protected void populateXObjectDictionaryForIndexColorModel(PDFDictionary dict,
IndexColorModel icm) {
// code deleted
Integer index = getIndexOfFirstTransparentColorInPalette(icm);
if (index != null) {
PDFArray mask = new PDFArray(dict);
mask.add(index);
mask.add(index);
dict.put("Mask", mask);
}
}
{code}
[https://github.com/apache/fop/blob/ff452f3685a02715aee791d651b58dd1797ddfd4/fop-core/src/main/java/org/apache/fop/render/pdf/AbstractImageAdapter.java]
h4. why care to support?
Libraries such as pngquant optimize images for web usage and low footprint and
can not be used directly with Apache FOP as source images:
{quote}"*pngquant*: Lossy PNG optimizer, reducing a PNG image down to an 8 bit
color palette with dithering. It will build indexed-color PNG's with alpha
transparency colors conveyed in the tRNS chunk."
{quote}
[https://imagemagick.org/Usage/formats/#png_non-im]
h4. sample data
{code}
!
{code}
was:
h4. TL;NR
When using palette-based transparency via tRNS chunk in PNG images, the
generated PDFs are broken.
h4. palette-based with transparency.
{quote}*8.5.2. Palette-Based with Transparency*
"The PNG spec forbids the use of a full alpha channel with palette-based
images, but it does allow `cheap alpha` via the transparency chunk, tRNS [...]
In effect, it transforms the palette from an RGB lookup table to an RGBA table
[...] a single PNG transparency entry should come first"{quote}
http://www.libpng.org/pub/png/book/chapter08.html#png.ch08.div.5.2
h4. general PDF encoding strategy
Option 1:
{quote}most libraries just undo the palettization and use a full color image +
mask.{quote}
https://forums.adobe.com/message/6148131#6148131
Option 2:
{quote}SMask would be simpler since you could leave the palette in place{quote}
https://forums.adobe.com/message/6148557#6148557
h4. what does FOP do
It creates a chroma-key mask:
{code}
/Type /XObject
/ColorSpace [/Indexed /DeviceRGB 255 <D9DB ...>]
/Mask [182 182 151 151 158 158]
{code}
Acrobat Reader even fails to read those files while PDF viewers embedded into
Chrome/Firefox read the file, but show visual glitches.
FOP doesn't implement Option 1 yet. None of the ImageLoaders is prepared to do
that:
https://wiki.apache.org/xmlgraphics-fop/HowTo/ImageLoaderRawPNG
In fact the following results in missing Option 2:
{code}
public void setup(PDFDocument doc) {
RenderedImage ri = getImage().getRenderedImage();
// code deleted
Raster raster = GraphicsUtil.getAlphaRaster(ri);
if (raster != null) {
AlphaRasterImage alphaImage = new AlphaRasterImage("Mask:" + getKey(),
raster);
this.softMask = doc.addImage(null, alphaImage).makeReference();
}
{code}
https://github.com/apache/fop/blob/ff452f3685a02715aee791d651b58dd1797ddfd4/fop-core/src/main/java/org/apache/fop/render/pdf/ImageRenderedAdapter.java
In the above code raster will be null and therefore FOP fails to initialize a
softMask. {{GraphicsUtil.getAlphaRaster()}} calls internally
{{ColorModel.getAlphaRaster()}}, which returns null:
{quote}If this is an IndexColorModel which has alpha in the lookup table, this
method will return null since there is no spatially discrete alpha
channel.{quote}
https://docs.oracle.com/javase/8/docs/api/index.html?java/awt/image/ColorModel.html
This code will then add the chroma key mask to the image:
{code}
protected void populateXObjectDictionaryForIndexColorModel(PDFDictionary dict,
IndexColorModel icm) {
// code deleted
Integer index = getIndexOfFirstTransparentColorInPalette(icm);
if (index != null) {
PDFArray mask = new PDFArray(dict);
mask.add(index);
mask.add(index);
dict.put("Mask", mask);
}
}
{code}
https://github.com/apache/fop/blob/ff452f3685a02715aee791d651b58dd1797ddfd4/fop-core/src/main/java/org/apache/fop/render/pdf/AbstractImageAdapter.java
h4. why care to support?
Libraries such as pngquant optimize images for web usage and low footprint and
can not be used directly with Apache FOP as source images:
{quote}"*pngquant*: Lossy PNG optimizer, reducing a PNG image down to an 8 bit
color palette with dithering. It will build indexed-color PNG's with alpha
transparency colors conveyed in the tRNS chunk."{quote}
https://imagemagick.org/Usage/formats/#png_non-im
> Support palette-based transparency via tRNS chunk of PNG images in PDF export
> -----------------------------------------------------------------------------
>
> Key: FOP-2847
> URL: https://issues.apache.org/jira/browse/FOP-2847
> Project: FOP
> Issue Type: Bug
> Components: renderer/pdf
> Affects Versions: 2.3
> Reporter: Tobias Hain
> Priority: Major
> Attachments: sample_image_palette_tRNS.png
>
>
> h4. TL;NR
> When using palette-based transparency via tRNS chunk in PNG images, the
> generated PDFs are broken.
> h4. palette-based with transparency.
> {quote}*8.5.2. Palette-Based with Transparency*
> "The PNG spec forbids the use of a full alpha channel with palette-based
> images, but it does allow `cheap alpha` via the transparency chunk, tRNS
> [...] In effect, it transforms the palette from an RGB lookup table to an
> RGBA table [...] a single PNG transparency entry should come first"
> {quote}
> [http://www.libpng.org/pub/png/book/chapter08.html#png.ch08.div.5.2]
> h4. general PDF encoding strategy
> Option 1:
> {quote}most libraries just undo the palettization and use a full color image
> + mask.
> {quote}
> [https://forums.adobe.com/message/6148131#6148131]
> Option 2:
> {quote}SMask would be simpler since you could leave the palette in place
> {quote}
> [https://forums.adobe.com/message/6148557#6148557]
> h4. what does FOP do
> It creates a chroma-key mask:
> {code:java}
> /Type /XObject
> /ColorSpace [/Indexed /DeviceRGB 255 <D9DB ...>]
> /Mask [182 182 151 151 158 158]
> {code}
> Acrobat Reader even fails to read those files while PDF viewers embedded into
> Chrome/Firefox read the file, but show visual glitches.
> FOP doesn't implement Option 1 yet. None of the ImageLoaders is prepared to
> do that:
> [https://wiki.apache.org/xmlgraphics-fop/HowTo/ImageLoaderRawPNG]
> In fact the following results in missing Option 2:
> {code:java}
> public void setup(PDFDocument doc) {
> RenderedImage ri = getImage().getRenderedImage();
> // code deleted
> Raster raster = GraphicsUtil.getAlphaRaster(ri);
> if (raster != null) {
> AlphaRasterImage alphaImage = new AlphaRasterImage("Mask:" + getKey(),
> raster);
> this.softMask = doc.addImage(null, alphaImage).makeReference();
> }
> {code}
> [https://github.com/apache/fop/blob/ff452f3685a02715aee791d651b58dd1797ddfd4/fop-core/src/main/java/org/apache/fop/render/pdf/ImageRenderedAdapter.java]
> In the above code raster will be null and therefore FOP fails to initialize a
> softMask. {{GraphicsUtil.getAlphaRaster()}} calls internally
> {{ColorModel.getAlphaRaster()}}, which returns null:
> {quote}If this is an IndexColorModel which has alpha in the lookup table,
> this method will return null since there is no spatially discrete alpha
> channel.
> {quote}
> [https://docs.oracle.com/javase/8/docs/api/index.html?java/awt/image/ColorModel.html]
> This code will then add the chroma key mask to the image:
> {code:java}
> protected void populateXObjectDictionaryForIndexColorModel(PDFDictionary
> dict, IndexColorModel icm) {
> // code deleted
> Integer index = getIndexOfFirstTransparentColorInPalette(icm);
> if (index != null) {
> PDFArray mask = new PDFArray(dict);
> mask.add(index);
> mask.add(index);
> dict.put("Mask", mask);
> }
> }
> {code}
> [https://github.com/apache/fop/blob/ff452f3685a02715aee791d651b58dd1797ddfd4/fop-core/src/main/java/org/apache/fop/render/pdf/AbstractImageAdapter.java]
> h4. why care to support?
> Libraries such as pngquant optimize images for web usage and low footprint
> and can not be used directly with Apache FOP as source images:
> {quote}"*pngquant*: Lossy PNG optimizer, reducing a PNG image down to an 8
> bit color palette with dithering. It will build indexed-color PNG's with
> alpha transparency colors conveyed in the tRNS chunk."
> {quote}
> [https://imagemagick.org/Usage/formats/#png_non-im]
> h4. sample data
> {code}
> !
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
