def rewrite_png(pike: Pdf, im_obj: Object,
compdata) -> None: # pragma: no cover
# When a PNG is inserted into a PDF, we more or less copy the IDAT section from
# the PDF and transfer the rest of the PNG headers to PDF image metadata.
# One thing we have to do is tell the PDF reader whether a predictor was used
# on the image before Flate encoding. (Typically one is.)
# According to Leptonica source, PDF readers don't actually need us
# to specify the correct predictor, they just need a value of either:
# 1 - no predictor
# 10-14 - there is a predictor
# Leptonica's compdata->predictor only tells TRUE or FALSE
# 10-14 means the actual predictor is specified in the data, so for any
# number >= 10 the PDF reader will use whatever the PNG data specifies.
# In practice Leptonica should use Paeth, 14, but 15 seems to be the
# designated value for "optimal". So we will use 15.
# See:
# - PDF RM 7.4.4.4 Table 10
# - https://github.com/DanBloomberg/leptonica/blob/master/src/pdfio2.c#L757
predictor = 15 if compdata.predictor > 0 else 1
dparms = Dictionary(Predictor=predictor)
if predictor > 1:
dparms.BitsPerComponent = compdata.bps # Yes, this is redundant
dparms.Colors = compdata.spp
dparms.Columns = compdata.w
im_obj.BitsPerComponent = compdata.bps
im_obj.Width = compdata.w
im_obj.Height = compdata.h
log.debug(
f"PNG {im_obj.objgen}: palette={compdata.ncolors} spp={compdata.spp} bps={compdata.bps}"
)
if compdata.ncolors > 0:
# .ncolors is the number of colors in the palette, not the number of
# colors used in a true color image. The palette string is always
# given as RGB tuples even when the image is grayscale; see
# https://github.com/DanBloomberg/leptonica/blob/master/src/colormap.c#L2067
palette_pdf_string = compdata.get_palette_pdf_string()
palette_data = pikepdf.Object.parse(palette_pdf_string)
palette_stream = pikepdf.Stream(pike, bytes(palette_data))
palette = [
Name.Indexed, Name.DeviceRGB, compdata.ncolors - 1, palette_stream
]
cs = palette
else:
# ncolors == 0 means we are using a colorspace without a palette
if compdata.spp == 1:
cs = Name.DeviceGray
elif compdata.spp == 4:
cs = Name.DeviceCMYK
else: # spp == 3
cs = Name.DeviceRGB
im_obj.ColorSpace = cs
im_obj.write(compdata.read(), filter=Name.FlateDecode, decode_parms=dparms)
def transcode_pngs(pike, images, image_name_fn, root, log, options):
if options.optimize >= 2:
png_quality = (
max(10, options.png_quality - 10),
min(100, options.png_quality + 10),
)
with concurrent.futures.ThreadPoolExecutor(
max_workers=options.jobs) as executor:
for xref in images:
log.debug(image_name_fn(root, xref))
executor.submit(
pngquant.quantize,
image_name_fn(root, xref),
png_name(root, xref),
png_quality[0],
png_quality[1],
)
for xref in images:
im_obj = pike.get_object(xref, 0)
try:
compdata = leptonica.CompressedData.open(png_name(root, xref))
except leptonica.LeptonicaError as e:
# Most likely this means file not found, i.e. quantize did not
# produce an improved version
log.error(e)
continue
# If re-coded image is larger don't use it - we test here because
# pngquant knows the size of the temporary output file but not the actual
# object in the PDF
if len(compdata) > int(im_obj.stream_dict.Length):
log.debug(
f"pngquant: pngquant did not improve over original image "
f"{len(compdata)} > {int(im_obj.stream_dict.Length)}")
continue
# When a PNG is inserted into a PDF, we more or less copy the IDAT section from
# the PDF and transfer the rest of the PNG headers to PDF image metadata.
# One thing we have to do is tell the PDF reader whether a predictor was used
# on the image before Flate encoding. (Typically one is.)
# According to Leptonica source, PDF readers don't actually need us
# to specify the correct predictor, they just need a value of either:
# 1 - no predictor
# 10-14 - there is a predictor
# Leptonica's compdata->predictor only tells TRUE or FALSE
# From there the PNG decoder can infer the rest from the file.
# In practice the predictor should be Paeth, 14, so we'll use that.
# See:
# - PDF RM 7.4.4.4 Table 10
# - https://github.com/DanBloomberg/leptonica/blob/master/src/pdfio2.c#L757
predictor = 14 if compdata.predictor > 0 else 1
dparms = Dictionary(Predictor=predictor)
if predictor > 1:
dparms.BitsPerComponent = compdata.bps # Yes, this is redundant
dparms.Colors = compdata.spp
dparms.Columns = compdata.w
im_obj.BitsPerComponent = compdata.bps
im_obj.Width = compdata.w
im_obj.Height = compdata.h
if compdata.ncolors > 0:
# .ncolors is the number of colors in the palette, not the number of
# colors used in a true color image
palette_pdf_string = compdata.get_palette_pdf_string()
palette_data = pikepdf.Object.parse(palette_pdf_string)
palette_stream = pikepdf.Stream(pike, bytes(palette_data))
palette = [
Name.Indexed,
Name.DeviceRGB,
compdata.ncolors - 1,
palette_stream,
]
cs = palette
else:
if compdata.spp == 1:
# PDF interprets binary-1 as black in 1bpp, but PNG sets
# black to 0 for 1bpp. Create a palette that informs the PDF
# of the mapping - seems cleaner to go this way but pikepdf
# needs to be patched to support it.
# palette = [Name.Indexed, Name.DeviceGray, 1, b"\xff\x00"]
# cs = palette
cs = Name.DeviceGray
elif compdata.spp == 3:
cs = Name.DeviceRGB
elif compdata.spp == 4:
cs = Name.DeviceCMYK
if compdata.bps == 1:
im_obj.Decode = [
1, 0
] # Bit of a kludge but this inverts photometric too
im_obj.ColorSpace = cs
im_obj.write(compdata.read(),
filter=Name.FlateDecode,
decode_parms=dparms)
def transcode_pngs(pike, images, image_name_fn, root, log, options):
if options.optimize >= 2:
png_quality = (
max(10, options.png_quality - 10),
min(100, options.png_quality + 10),
)
with concurrent.futures.ThreadPoolExecutor(
max_workers=options.jobs
) as executor:
for xref in images:
log.debug(image_name_fn(root, xref))
executor.submit(
pngquant.quantize,
image_name_fn(root, xref),
png_name(root, xref),
png_quality[0],
png_quality[1],
)
for xref in images:
im_obj = pike.get_object(xref, 0)
try:
compdata = leptonica.CompressedData.open(png_name(root, xref))
except leptonica.LeptonicaError as e:
# Most likely this means file not found, i.e. quantize did not
# produce an improved version
log.error(e)
continue
# If re-coded image is larger don't use it - we test here because
# pngquant knows the size of the temporary output file but not the actual
# object in the PDF
if len(compdata) > int(im_obj.stream_dict.Length):
log.debug(
f"pngquant: pngquant did not improve over original image "
f"{len(compdata)} > {int(im_obj.stream_dict.Length)}"
)
continue
# When a PNG is inserted into a PDF, we more or less copy the IDAT section from
# the PDF and transfer the rest of the PNG headers to PDF image metadata.
# One thing we have to do is tell the PDF reader whether a predictor was used
# on the image before Flate encoding. (Typically one is.)
# According to Leptonica source, PDF readers don't actually need us
# to specify the correct predictor, they just need a value of either:
# 1 - no predictor
# 10-14 - there is a predictor
# Leptonica's compdata->predictor only tells TRUE or FALSE
# From there the PNG decoder can infer the rest from the file.
# In practice the predictor should be Paeth, 14, so we'll use that.
# See:
# - PDF RM 7.4.4.4 Table 10
# - https://github.com/DanBloomberg/leptonica/blob/master/src/pdfio2.c#L757
predictor = 14 if compdata.predictor > 0 else 1
dparms = Dictionary(Predictor=predictor)
if predictor > 1:
dparms.BitsPerComponent = compdata.bps # Yes, this is redundant
dparms.Colors = compdata.spp
dparms.Columns = compdata.w
im_obj.BitsPerComponent = compdata.bps
im_obj.Width = compdata.w
im_obj.Height = compdata.h
if compdata.ncolors > 0:
# .ncolors is the number of colors in the palette, not the number of
# colors used in a true color image
palette_pdf_string = compdata.get_palette_pdf_string()
palette_data = pikepdf.Object.parse(palette_pdf_string)
palette_stream = pikepdf.Stream(pike, bytes(palette_data))
palette = [
Name.Indexed,
Name.DeviceRGB,
compdata.ncolors - 1,
palette_stream,
]
cs = palette
else:
if compdata.spp == 1:
# PDF interprets binary-1 as black in 1bpp, but PNG sets
# black to 0 for 1bpp. Create a palette that informs the PDF
# of the mapping - seems cleaner to go this way but pikepdf
# needs to be patched to support it.
# palette = [Name.Indexed, Name.DeviceGray, 1, b"\xff\x00"]
# cs = palette
cs = Name.DeviceGray
elif compdata.spp == 3:
cs = Name.DeviceRGB
elif compdata.spp == 4:
cs = Name.DeviceCMYK
if compdata.bps == 1:
im_obj.Decode = [1, 0] # Bit of a kludge but this inverts photometric too
im_obj.ColorSpace = cs
im_obj.write(compdata.read(), filter=Name.FlateDecode, decode_parms=dparms)