Dominik Schmidt created TIKA-4861:
-------------------------------------
Summary: Content-based detection for TIFF-based raw camera
formats (NEF, PEF, ARW, DNG, SRW) and for RAF, RW2 and MRW
Key: TIKA-4861
URL: https://issues.apache.org/jira/browse/TIKA-4861
Project: Tika
Issue Type: Improvement
Reporter: Dominik Schmidt
Most raw camera formats are still detected by file extension only. Without a
name, a stream is detected as image/tiff (or application/octet-stream for the
non-TIFF ones), parsed by the generic TIFF parser, and the previews
RawTiffParser emits as THUMBNAIL embedded documents (TIKA-4824, TIKA-4851) are
never seen. Clients that upload without a reliable file name, such as
tika-server callers sending a bare body, get no thumbnail for these files.
Two groups:
1. Formats with a fixed signature that tika-mimetypes.xml does not match yet:
Fuji RAF starts with "FUJIFILMCCD-RAW", Panasonic RW2 with "IIU\0" (a TIFF
variant with its own magic number 0x55), Minolta MRW with "\0MRM"; Olympus ORF
also occurs big-endian as "MMOR" and as "IIRS" next to the "IIRO" we match.
These are magic entries only. Canon CR2/CR3 already have theirs (TIKA-3991).
2. Formats that are valid TIFF files: Nikon NEF/NRW, Pentax PEF/PTX, Sony
ARW/SRF/SR2, Samsung SRW and Adobe DNG. No fixed-offset magic can tell them
from a plain TIFF, and the Make tag alone is not enough: Nikon and Pentax
cameras also write real TIFFs with their name in Make. ExifTool identifies
these by content in two steps: a DNGVersion tag (0xC612) means DNG, and a
vendor-specific Compression value (0x0103) in any IFD names the format (34713
NEF, 32767 ARW, 65535 PEF, 32770 SRW). That misses uncompressed raws
(Compression 1), which can be caught by the image data itself: a raw holds an
image with PhotometricInterpretation 32803 (CFA) or 34892 (LinearRaw) in IFD0
or a SubIFD (0x014A), where a plain TIFF has 2 (RGB); there Make picks the
vendor.
Proposal: a Detector in tika-parser-image-module, registered via SPI, that only
acts when the TIFF magic matches and reads the IFD chain and SubIFDs with a
bounded reader (the one in RawTiffParser already handles classic TIFF and
BigTIFF), applying those rules in that order and returning
application/octet-stream otherwise so image/tiff stands. A caller-supplied type
or file name keeps precedence as usual.
Both parts can go into one PR; the second is where the review effort is.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
