TiffSaver: produce valid 4-channel RGBA images #4059
Conversation
CGDogan
changed the title
|
For reference, the original logic was introduced in #3057. It would help to know which viewer(s) can't handle TIFFs converted without this change, so that we can better evaluate the behavior of this change across a range of use cases. |
|
My mac can’t, but the reason I’m concerned with this is that I bfconvert images to tiff so that I can view with OpenSlide, but OpenSlide then shows it as monochrome. But for comparison, I searched for “view tiff online” and opened 5 different sites, none of them could view it. Here's the file: https://github.com/ome/bioformats/files/11995909/out5.tiff.zip |
|
This pull request has been mentioned on Image.sc Forum. There might be relevant details there: https://forum.image.sc/t/bfconvert-breaks-images-with-alpha-layer-wrong-interleave/83482/5 |
There was a problem hiding this comment.
Running bfconvert with an RGBA input file and test.tiff as the output file name does result in a 4-channel RGB TIFF which can be displayed with showinf, ImageMagick's display, and Windows Photos. However, display shows this warning:
display-im6.q16: Sum of Photometric type-related color channels and ExtraSamples doesn't match SamplesPerPixel. Defining non-color channels as ExtraSamples.. `TIFFReadDirectory' @ warning/tiff.c/TIFFWarnings/912.
The ExtraSamples tag should be set in this case, with a single value (likely 0). See pages 31-32 of the TIFF specification.
|
It should work now? |
There was a problem hiding this comment.
This looks fine to me now. If I convert a 4-channel RGBA PNG to TIFF with bfconvert, the resulting TIFF has the PhotometricInterpretation, BitsPerSample, and ExtraSamples values that are expected. Opening the file with ImageMagick's display command shows a correct image with no warnings.
Converting the same file with bfconvert -separate continues to produce 4 separate IFDs, none of which have an ExtraSamples tag. The PhotometricInterpretation in this case is 1 (BLACK_IS_ZERO) and BitsPerSample has one entry, all as expected.
As I asked on https://forum.image.sc/t/bfconvert-breaks-images-with-alpha-layer-wrong-interleave/83482, bfconvert, when given 4-channel images, outputs with the monochrome tag. This means that many image decoders will try to read it as monochrome and show an invalid image. In general, we should save 4-channel images like 3-channel ones, because RGBA tagged as RGB is a supported configuration in the TIFF specification (https://www.itu.int/itudoc/itu-t/com16/tiff-fx/docs/tiff6.pdf pg 77), where even the order is specified: R,G,B, then A. This will result in BioFormats producing much more compatible tiff files.
EDIT: Rebased, considering that the TIFF specification does not limit the additional alpha channel to RGB, but it limits its premultipliedness property only to RGB. As such, YCbCr can also have an alpha channel