Fix TIFF processing. Add tests to prevent regression in OCR for gif, jpg, jp2, tiff, webp #587
Conversation
ingestors/media/tiff.py
Outdated
| @@ -23,7 +23,8 @@ def ingest(self, file_path, entity): | |||
| entity.schema = model.get("Pages") | |||
| pdf_path = self.make_work_file("tiff.pdf") | |||
| self.exec_command( | |||
| "tiff2pdf", file_path, "-x", "300", "-y", "300", "-o", pdf_path | |||
| "tiff2pdf", file_path, "-n", "-j", "-x", "300", "-y", "300", "-o", pdf_path | |||
There was a problem hiding this comment.
Could you explain why TIFF ingestion was previously failing with these to options enabled? It’s not obvious to me what they had an impact and why removing them fixes the issue.
There was a problem hiding this comment.
Also, did you figure out why only some TIFF files were affected and others not?
There was a problem hiding this comment.
@tillprochaska added a thorough explanation in the PR description, let me know if it is informative
There was a problem hiding this comment.
Thanks for the detailed explanation, this was really helpful.
One follow up question: My understanding of TIFF is very limited, but as TIFF can have multiple pages etc. and apparently embed image data using other image formats, is it possible that a TIFF file contains both JPEG compressed image data as well as other data?
catileptic
changed the title
|
Just for future reference, not necessarily something we need to handle right now. But I found this issue in the libtiff repo which might describe a similar issue: https://gitlab.com/libtiff/libtiff/-/issues/13
If I understand the referenced issue correctly, it might be that it has been fixed in recent libtiff version. The version we’re using is quite old (4.1.0, released in 2019) |
This PR addresses the fact that some TIFF images were not being OCR'ed correctly.
This stemmed from the fact that, if the TIFF file contained data with JPEG compression, the
tiff2pdfcommand, as it existed before this commit, would generate a PDF with an empty image.To address this, we have added the
-nand-jflags to thetiff2pdfcommand:-ncommand results in the JPEG-compressed data actually being written to the PDF (according to the tiff2pdf man page this flag sets "no passthrough" option? - not 100% sure on this)-jflag sets the compression type and keeps the resulting PDF from being blown up in sizeNow, TIFF images that do not contain JPEG-compressed data are not converted correctly to PDF and error out when the
-jflag is used. Thus, this PR attempts to first use the-nand-jflags and then tries to run the command without them, in case it fails. This does not work the other way around, because, if there is JPEG-compressed data in the TIFF, thetiff2pdfcommand does not fail, but instead produces an empty image inside a PDF.This PR also adds a test that specifically checks a regression in the OCR behaviour described in this PR. It also updates the existing TIFF parsing test because TIFF images get converted into one "Pages" entity with several "Page" entities and it's only the "Pages" entity that has the "mimeType" property set, so the test specifically looks for that entity before asserting whether or not its properties contain the expected values.