[ 
https://issues.apache.org/jira/browse/TIKA-4776?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18100429#comment-18100429
 ] 

ASF GitHub Bot commented on TIKA-4776:
--------------------------------------

dschmidt opened a new pull request, #2981:
URL: https://github.com/apache/tika/pull/2981

   With metadata-extractor 2.21.0 on main the Apple maker note tags written as 
LONG8 (TIFF format code 16) come through, so this adds the regression test 
discussed in the ticket. The fixture is the still half of a real Apple Live 
Photo (iPhone 15 Pro, iOS 18.5) from the MIT licensed osxphotos test suite; the 
test asserts the content identifier and the Live Photo ID (maker note tag 
0x0017, value 5283876), which metadata-extractor 2.20.0 silently dropped. With 
this in place TIKA-4776 can be resolved.
   




> Apple maker note tags written as LONG8 are dropped
> --------------------------------------------------
>
>                 Key: TIKA-4776
>                 URL: https://issues.apache.org/jira/browse/TIKA-4776
>             Project: Tika
>          Issue Type: Improvement
>            Reporter: Dominik Schmidt
>            Priority: Major
>
> Apple maker notes in HEIC/JPEG images store several tags as LONG8 (TIFF
> format code 16), e.g. the Live Photo video index (tag 0x0017). Example:
> for a Live Photo still taken with an iPhone 15 Pro on iOS 18.5, tag
> 0x0017 is written with format code 16 and value 5283876.
> The bundled metadata-extractor (2.20.0) only decodes the classic TIFF
> format codes 1-12 and drops such tags, so Tika never emits
> "img:Live Photo ID" even though metadata-extractor's
> AppleMakernoteDirectory defines the tag name. Only the content
> identifier ("img:Content Identifier", stored as a plain string) comes
> through today.
> This matters for pairing Apple Live Photos: the still image and the
> QuickTime video share a content identifier (the video side is covered
> by TIKA-2861), and the Live Photo video index is the image-side marker
> that reliably identifies a still as part of a Live Photo (whereas just 
> "Content Identifier" seems to be a bit weak).
> This is fixed upstream in metadata-extractor:
> [https://github.com/drewnoakes/metadata-extractor/issues/738]
> [https://github.com/drewnoakes/metadata-extractor/pull/739]
> Verified end-to-end with a patched metadata-extractor build: Tika then
> emits both "img:Content Identifier" and "img:Live Photo ID" for a real
> Live Photo HEIC, with no Tika code change required.
> Once a metadata-extractor release containing the fix is available, the
> dependency (metadata.extractor.version in tika-parent) needs a bump.
> It would be great to add a regression test then.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to