ICC profiles

[telenkor]

Please tell me, сan we work with ICC color profiles? I'm interested in adding a profile to image. If this is not possible, please help me understand the situation:
There is a raw picture, let's call it: «source_raw».
There is a RGB picture, let's call it: «rgb_target».
«rgb_target» currently has the ICC profile, but does not have other metadata.

        image = exiv2.ImageFactory.open(rgb_target)
        image.setExifData(exif_from_source_raw)
        image.writeMetadata()

This does only one good thing - it adds metadata from source_raw to rgb_target, but at the same time it destroys the existing ICC-profile. I understand that it replaces the old with the new. Is it possible to somehow combine existing ICC data and new other data? (e.g. image.setExifData(exif_from_source_raw + exif_from_rgb_target))

[jim-easterbrook]

The Exiv2::Image class has iccProfile() and setIccProfile() methods which should be accessible from Python.
https://exiv2.org/doc/classExiv2_1_1Image.html#a7f0c33da949b99bcf0b6c63d47ca6f33

[jim-easterbrook]

As I said earlier:

You should be able to pass some data to the 3rd constructor, but this is not currently callable from Python. I'll fix that in the next release of python-exiv2.

You can't currently copy data into an Exiv2::DataBuf. Please wait a few days for me to do a new release with a fix. I've assigned it to bug #15.

[jim-easterbrook]

I've made some more changes to the Python interface to Exiv2::DataBuf this morning. In the next release the data() method returns a Python memoryview of the data which is writeable. This means you will be able to modify ICC profiles in place without needing to create new exiv2.DataBuf objects.

[telenkor]

Thank you very much! I'll be looking forward to the release of the build on pypi.

[jim-easterbrook]

I'm uploading v0.15.0 to PyPI now. I think it has everything you need to work with ICC profiles.

Thank you for reporting the bugs you've found and starting some discussions. They've made me look much closer at how some things work, which has led to some worthwhile improvements in python-exiv2.

[telenkor]

Great news at the end of the week! I'll be exploring the new version this weekend. I am glad that I helped make my tiny contribution to the development of the project.

[telenkor]

Everything works fine: modification metadata, embedding icc (for Jpeg and Tiff). But the WebP format came as a surprise. Metadata is good. But when embedding an ICC profile, the file becomes unreadable. A check in the HEX editor showed that the EXIV2 does not put down the necessary marker "ICCP" WebP Container Specification. Everything is OK on the CLI version 0.27.7 of Windows 😕

I manually place the desired marker and the image comes to life. Of course, as long as the data is in RAM, I can patch the file. However, this behavior in relation to this format seems strange. Maybe there was an error somewhere or some factor was missed?

Here is a slightly simplified code from my project regarding working with EXIV2. If you comment out the # Embedd ICC profile line, everything is fine with the file.

from io import BytesIO

from PIL import Image
import exiv2  # v0.15

color_space = None
# Initialize common ICC profiles (sRGB IEC61966-2.1 and Adobe RGB 1998) in buffer
with open("/tools/srgb.icc", "rb") as file_icc:
    icc_data_srgb = file_icc.read()
with open("/tools/adobe_rgb_1998.icc", "rb") as file_icc:
    icc_data_argb98 = file_icc.read()

input_file = "/Users/test/Desktop/0050-master-flat.tif"
img = Image.open(input_file)
img.thumbnail((900, 900), Image.Resampling.BICUBIC)

buffer_ = BytesIO()
# img.save(buffer_, format="JPEG", subsampling=1, progressive=1, optimize=1, quality=75)
img.save(buffer_, format="webp", method=6, quality=75)

image_ref_file = exiv2.ImageFactory.open(input_file)
image_ref_file.readMetadata()
exif_ref = image_ref_file.exifData()

i = exif_ref.begin()
end = exif_ref.end()
exif_custom = exiv2.ExifData()
while i != end:
    if "ColorSpace" in i.key():
        # 1 - sRGB, 2 - Adobe RGB 1998
        if i.toString() == "1":
            color_space = 1
        elif i.toString() == "2":
            color_space = 2
    if i.key() in (
            'Exif.Photo.ExposureTime',
            'Exif.Photo.ISOSpeedRatings',
            'Exif.Image.Make'
    ):
        exif_custom[i.key()] = i.value()
    next(i)

image = exiv2.ImageFactory.open(buffer_.getvalue())
image.setExifData(exif_custom)

# Embedd ICC profile
image.setIccProfile(exiv2.DataBuf(icc_data_srgb if color_space == 1 else icc_data_argb98))

image.writeMetadata()

io = image.io()
# with open("/Volumes/Disk/out/0050-master-flat.jpg", 'wb') as file:
#     file.write(io.mmap())
with open("/Volumes/Disk/out/0050-master-flat.webp", 'wb') as file:
    file.write(io.mmap())
io.munmap()
io.close()

Tiff and profiles
P.S. If a valid WebP does not open anywhere, then open it in a browser (Chrome, Mozilla and others).

[jim-easterbrook]

That's a bug in the libexiv2 webp implementation: Exiv2/exiv2#2734

When I get libexiv2 0.28.1 (or later) properly working with python-exiv2 then it should be fixed.

[telenkor]

This is not a critical issue. I made a patch to fix it. Hopefully the new library design will be usable with python-exiv2 soon.

        # Patch ICCP chunk header https://github.com/Exiv2/exiv2/issues/2734
        image_content = bytearray(io.mmap())
        if self.preview_compression_format == "webp":
            wrong_header = image_content.index(b"\x00\x56\x50\x38\x58")
            right_header = b"\x00\x49\x43\x43\x50"
            image_content[wrong_header:wrong_header + 5] = right_header

[jim-easterbrook]

Your snippet shows an mmap() without a corresponding munmap(). It's easier to use the io object's buffer interface to initialise the bytearray: image_content = bytearray(io).

You might be able to do an in-place modification by using a memoryview of the io object.

        # Patch ICCP chunk header https://github.com/Exiv2/exiv2/issues/2734
        if self.preview_compression_format == "webp":
            with memoryview('io') as ptr:
                wrong_header = bytes(ptr).index(b"\x00\x56\x50\x38\x58")
                right_header = b"\x00\x49\x43\x43\x50"
                ptr[wrong_header:wrong_header + 5] = right_header
        image_content = bytearray(io)

[telenkor]

Thanks, let's try this.