JPEG compression support

[johnnychen94]

I found this "quad-jpeg.tif" from the libtiff test images https://libtiff.gitlab.io/libtiff/images.html

I just put up a draft JpegTurbo version to support abbreviated data stream JuliaIO/JpegTurbo.jl#25, I believe jpeg_decode(table_bytes, data_bytes) should make things doable. I'd be glad to support any JpegTurbo features needed here.

Because ColorTypes doesn't support YCbCr{N0f8}, we might need some special trick to work around this.

quad-jpeg.tif.zip

[tlnagy]

I don't see myself having the bandwidth this year for this (trying to wrap up my PhD work right now). Happy to review a PR though.

What would be involved with adding YCbCr{N0f8} to colortypes?

[johnnychen94]

Oh, I then have to throw myself into another image format then :) I guess that won't happen until Aug.

What would be involved with adding YCbCr{N0f8} to colortypes?

That probably won't fix the problem. -- Which colorspace does JPEG use to compress the data doesn't affect the output type that much, IIUC, libjpeg-turbo always outputs one of Gray, RGB, AGray, ARGB color. Thus we can't use the current interpretation for YCbCr to allocate the output array.

TiffImages.jl/src/layout.jl

Line 42 in e9cc707

interpretation(::Val{PHOTOMETRIC_YCBCR}) = YCbCr

Is YCbCr used somewhere else?

[tlnagy]

Sorry for the delay here, but interpretation is only used for the mapping of the reported colorway in the TIFF file to the julia equivalent, e.g. if there are 3 Float64s in a row then are they RGB{Float64}s or HSL{Float64}, etc. So that's the only place I explicitly use the ColorType structs.

I'm assuming this is the error you're talking about?

julia> TiffImages.load("/home/tlnagy/Downloads/libtiffpic/quad-jpeg.tif")
ERROR: TypeError: in YCbCr, in T, expected T<:AbstractFloat, got Type{FixedPointNumbers.N0f8}
Stacktrace:
  [1] getcache(ifd::TiffImages.IFD{UInt32})
    @ TiffImages ~/.julia/packages/TiffImages/P86VX/src/layout.jl:92
  [2] load(tf::TiffFile{UInt32, FileIO.Stream{FileIO.DataFormat{:TIFF}, IOStream, String}}, ifds::Vector{TiffImages.IFD{UInt32}}, ::Nothing; verbose::Bool)
    @ TiffImages ~/.julia/packages/TiffImages/P86VX/src/load.jl:66
  [3] load(tf::TiffFile{UInt32, FileIO.Stream{FileIO.DataFormat{:TIFF}, IOStream, String}}; verbose::Bool, mmap::Bool)
    @ TiffImages ~/.julia/packages/TiffImages/P86VX/src/load.jl:31
  [4] load(io::IOStream; verbose::Bool, mmap::Bool)
    @ TiffImages ~/.julia/packages/TiffImages/P86VX/src/load.jl:16
  [5] #10
    @ ~/.julia/packages/TiffImages/P86VX/src/load.jl:12 [inlined]
  [6] open(f::TiffImages.var"#10#11"{Bool, Bool}, args::String; kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Base ./io.jl:330
  [7] open
    @ ./io.jl:328 [inlined]
  [8] #load#9
    @ ~/.julia/packages/TiffImages/P86VX/src/load.jl:11 [inlined]
  [9] load(filepath::String)
    @ TiffImages ~/.julia/packages/TiffImages/P86VX/src/load.jl:11
 [10] top-level scope
    @ REPL[2]:1

Lets look at the first IFD in the TIFF file using TiffImages:

julia> io = open("/home/tlnagy/Downloads/libtiffpic/quad-jpeg.tif")
IOStream(<file /home/tlnagy/Downloads/libtiffpic/quad-jpeg.tif>)

julia> tf = read(io, TiffImages.TiffFile);

julia> ifd = first(tf); # get first ifd

julia> TiffImages.load!(tf, ifd); # load remote data

julia> ifd
IFD, with tags: 
	Tag(IMAGEWIDTH, 512)
	Tag(IMAGELENGTH, 384)
	Tag(BITSPERSAMPLE, UInt16[8, 8, 8])
	Tag(COMPRESSION, COMPRESSION_JPEG)
	Tag(PHOTOMETRIC, 6)
	Tag(STRIPOFFSETS, UInt32[8, 175, 342, 633, 2227, ...])
	Tag(SAMPLESPERPIXEL, 3)
	Tag(ROWSPERSTRIP, 16)
	Tag(STRIPBYTECOUNTS, UInt32[167, 167, 291, 1594, 1456, ...])
	Tag(PLANARCONFIG, 1)
	Tag(XPOSITION, 0x00000001//0x00000000)
	Tag(YPOSITION, 0x00000001//0x00000000)
	Tag(JPEGTABLES, Any[255, 216, 255, 219, 0, ...])
	Tag(REFERENCEBLACKWHITE, Rational{UInt32}[0x00000001//0x00000000, 0x00200000//0x1fe00000, 0x00200000//0x10000000, 0x00200000//0x1fe00000, 0x00200000//0x10000000, ...])

A photometric representation of 6 is, according to https://www.awaresystems.be/imaging/tiff/tifftags/photometricinterpretation.html, equal to YCbCr. Which means that we need to report the final image (after decompression) as YCbCr{N0f8} (BITSPERSAMPLE is 8) so the lack of YCbCr{N0f8} in Julia will prevent us from representing the results faithfully even if decompression support is added.

[MendeBadra]

---

Hello! I've been trying to open a large TIFF file using TiffImages.jl, but I encountered this error:

julia> img_path = "/media/jaki-73/HDD/MendePythonTest/Liver Fat WSI/1_20240522130835.tif";
julia> load(img_path)
Errors encountered while load File{DataFormat{:TIFF}, String}("/media/jaki-73/HDD/MendePythonTest/Liver Fat WSI/1_20240522130835.tif").
All errors:
===========================================
TaskFailedException

    nested task error: Compression COMPRESSION_JPEG is not implemented. Please open an issue against TiffImages.jl.

This is followed by a lengthy stack trace.

Additionally, I ran a code snippet similar to the previous comment:

julia> using TiffImages

julia> tf = read(io, TiffImages.TiffFile);

julia> ifd = first(tf)
IFD, with tags: 
    Tag(IMAGEWIDTH, 121600)
    Tag(IMAGELENGTH, 86272)
    Tag(BITSPERSAMPLE, UInt16[8, 8, 8])
    Tag(COMPRESSION, COMPRESSION_JPEG)
    Tag(PHOTOMETRIC, 2)
    Tag(SAMPLESPERPIXEL, 3)
    Tag(ROWSPERSTRIP, 86272)
    Tag(XRESOLUTION, 0x4b000000//0x00200000)
    Tag(YRESOLUTION, 0x4b000000//0x00200000)
    Tag(PLANARCONFIG, 1)
    Tag(TILEWIDTH, 256)
    Tag(TILELENGTH, 256)
    Tag(TILEOFFSETS, REMOTE@4515844682 UInt64[] len=160075)
    Tag(TILEBYTECOUNTS, REMOTE@4517125282 UInt64[] len=160075)
    Tag(JPEGTABLES, REMOTE@4518405882 Any[] len=289)
    Tag(YCBCRSUBSAMPLING, UInt16[2, 2])

julia> TiffImages.load!(tf, ifd)

It seems that TiffImages.jl does not support the JPEG compression method (COMPRESSION_JPEG), which caused the initial error. Could someone confirm if this is correct? If so, when can we expect support for this feature? As a beginner in Julia, I'm not sure about the effort required to implement this, so please forgive me if my question seems naive.

---