looking for deep color image formats/corpus, not too weird

[Paul W.]

I'm looking to find or collect a corpus of color images in more-than-8-bits per color, in some reasonably straightforward file format(s), e.g., 30-bit color with 10 bits per channel, packed 10/10/10 into 32 bits with 2 bits left over for padding or alpha, or 12/12/8 RGB, or 16-bits/channel 48-bit RGB or 64-bit RGBA, or whatever.

My actual goal is to figure out how to detect and compress various in-memory representations of images, like some of the weird deep-color representations natively supported by some graphics cards (10 to 12 bits per channel or whatever, maybe with different numbers of bits per channel).

(For the moment, I'm not as interested in really weird, camera-specific "raw" representations of photographs that depend on bizarre details of raw nonuniform RGB pixel layouts (e.g., where half the subpixels are green) though in principle I want to be able to handle that stuff too, eventually.)

I'm rather more interested in intermediate representations that may be used by hi-res and HDR image processing programs (or such modes of Photoshop/Gimp/photo or cinema editing software/whatever), such as 48-bit pixels with 16-bit color channels, or asymmetric monstrosities directly supported by the strange new instructions some processors or GPUs have. (E.g., the ones with one shared exponent for 3 mantissas, on the assumption that they're all on the same general scale that matters, because hey, it's probably a pixel where the relevant logarithmic intensity range is the same.)

What I'm really trying to do is to make sure I can discriminate between and various likely in-memory representations of images that are likely to be common in the near future, if not now, in programs that process images, and to make sure my code can figure out basic stride and alignment issues---e.g., that pixels are 24 vs. 32 vs. 36 vs. 40 vs. 48 bits---and where the channels are. (E.g., whether 32 bits is 10/10/10/2 RGBA, or 12/12/8 RGB.)

Any pointers to relevant test images, file formats, and image conversion utilities would be appreciated. (Anything particularly useful will mostly end up in free software, assuming I get that far.)

[SvenBent]

http://www.imagecompression.info/test_images/

16bits channels pictures both true colors+ and rgb+

[thorfdbg]

I'm looking to find or collect a corpus of color images in more-than-8-bits per color, in some reasonably straightforward file format(s), e.g., 30-bit color with 10 bits per channel, packed 10/10/10 into 32 bits with 2 bits left over for padding or alpha, or 12/12/8 RGB, or 16-bits/channel 48-bit RGB or 64-bit RGBA, or whatever.

Certainly. The JPEG has a good selection of images, most of them available under a permissive release. We currently have "intermediate dynamic range images" with bitdepths of 16 bits per sample stored in scRGB and encoded in PPM, plus the "raw" versions of them, and a couple of "high dynamic range" images using a floating point "randiance" type of conversion. These images are encoded in "pfm" and "openexr". I could also offer a conversion and measurement tool for many image formats under GPLv3 licence if you could make use of it. It speaks ppm,pgm,pbm,tif,png,exr,pgx,bmp and a couple of "raw" representations without a header. If interested, please sent me a pm and I can make the corpus available (to the extend the releases allow this). (Hopefully, the new jpeg.org websites will go online soon....)

[Paul W.]

Certainly. The JPEG has a good selection of images, most of them available under a permissive release. We currently have "intermediate dynamic range images" with bitdepths of 16 bits per sample stored in scRGB and encoded in PPM, plus the "raw" versions of them, and a couple of "high dynamic range" images using a floating point "randiance" type of conversion. These images are encoded in "pfm" and "openexr".

That sounds cool. For the moment I will fiddle with some of the pictures Sven pointed me to, but after that I will follow up on this.

I could also offer a conversion and measurement tool for many image formats under GPLv3 licence if you could make use of it. It speaks ppm,pgm,pbm,tif,png,exr,pgx,bmp and a couple of "raw" representations without a header. If interested, please sent me a pm and I can make the corpus available (to the extend the releases allow this). (Hopefully, the new jpeg.org websites will go online soon....)

That sounds great. Does your tool generate funny packed pixel formats like some graphics hardware uses, like 30-bit color in 4 bytes?

I'm particularly interested in in-memory image representations likely to be used now or soon inside programs like Photoshop and the Gimp, before final versions are written out as files, but not being a graphics guy, I don't really know what to expect. Any thoughts on that?

[nemequ]

I'm particularly interested in in-memory image representations likely to be used now or soon inside programs like Photoshop and the Gimp, before final versions are written out as files, but not being a graphics guy, I don't really know what to expect. Any thoughts on that?

For GIMP, see http://gegl.org/babl/#Pixel-formats. You might want to talk to the GIMP devs (#gimp on irc.gimp.org, or there are mailing lists if you prefer that) to get an idea of which of those formats are used the most often.

[thorfdbg]

That sounds great. Does your tool generate funny packed pixel formats like some graphics hardware uses, like 30-bit color in 4 bytes?

If you need that kind, the raw output formatter of the tool allows you to do even that. The syntax for the raw formatter is a bit crude, though, but this is due to the complexity of the problem.

I'm particularly interested in in-memory image representations likely to be used now or soon inside programs like Photoshop and the Gimp, before final versions are written out as files, but not being a graphics guy, I don't really know what to expect. Any thoughts on that?

Neither do I. I would expect that gimp and friends use a simple 48 bit RGB format internally for that, but I could be completely wrong. My tool uses a representation of a "bytes per pixel" and "bytes per row" indicator, plus a format per component. Thus, if you really need a 9-bit red, 7 bit green, 2 bit blue formatted in interleaved components - yes, can do.