[Barry Pearson]

Indeed, your experiment give some hint that raws aren't raws per say, it the file size isn't fixed. Interesting. Does it means the DNG could be read as a JPEG is the files header was modified to look like a JPEG?

No. A DNG file doesn't resemble a JPEG file in the slightest degree!

It means that "lossless JPEG" really, really, can compress data without loss. Repeat, without loss!

[funktionsfrei]

Structurally, DNG is a variant of TIFF. As are other RAW file formats (PEF, CR2 and whatsernames). TIFF is just a container, the data fields within it are what makes the difference.

The thing here is that Pentax uses lossless compression. That's nothing magical, but quite normal, as your computer's using ZIP, 7Z, RAR, DMG or whatever else kind of compressed file. All these compression variants are lossless, and so is the variant used by Pentax on their DNG and PEF files.

[Barry Pearson]

Structurally, DNG is a variant of TIFF. As are other RAW file formats (PEF, CR2 and whatsernames). TIFF is just a container, the data fields within it are what makes the difference.

The thing here is that Pentax uses lossless compression. That's nothing magical, but quite normal, as your computer's using ZIP, 7Z, RAR, DMG or whatever else kind of compressed file. All these compression variants are lossless, and so is the variant used by Pentax on their DNG and PEF files.

For interest, DNG is based on the ISO raw file format ISO 12234-2, also known as TIFF/EP. (Which in turn was based on TIFF).

ISO screwed up that standard, defining too little metadata and not making enough metadata mandatory. DNG adds in the extra metadata required. (ISO have been considering how to turn DNG into a revised ISO standard for years! They don't appear to be making much progress).

DNG uses "lossless JPEG" compression. I don't know what PEF uses. (Is that "Lossless JPEG" too?)

[rawr]

If not, then the raw is processed and can be any recipe chosen by the manufacturer to meet their goal.

Every stereo MP3 recorded at the same bit-rate and of 3 minutes duration is not an identical tune ...

---------- Post added 2016-06-07 at 08:06 AM ----------

all modern DSLRs have some level of raw-cooking

Indeed. If you look at the image data pipeline from lens into camera and into SD card, the data is 'cooked' by about 15 chips and codecs in the pipeline. Some data is even 'cooked' on-sensor.

Eg Exmor:


It's not a 'pure' data pathway by any means. It is electronics, after all.

---

Last edited by rawr; 06-06-2016 at 03:12 PM.

[biz-engineer]

Some data is even 'cooked' on-sensor.

Looks like there is some noise reduction on ADC bitstream post processing (decimation?) , also removes image details.

---------- Post added 07-06-16 at 08:31 ----------

It means that "lossless JPEG" really, really, can compress data without loss. Repeat, without loss!

Ah ah I know, but I honestly simply haven't reverse engineered a DNG, nor JPEG and I don't work for Ricoh, so I can be sure if it is lossless or not.

[beholder3]

I was wondering why the K-1 raw files are about 45 MB when the Nikon D800s files are 75MB...

Same Image and ISO:
http://www.dpreview.com/reviews/image-comparison?attr18=daylight&attr13_0=pe...dget=1&x=0&y=0

Pentax 36 Mpx: 42,7 MB
Nikon 36 Mpx: 74,4 MB
Canon 50 MPx: 63,9 MB
Sony 42 MPx: 41,4 MB

Sony uses the most aggressive compression obviously (even in a model with hopefully now lossless compression).
Canon seems to use similar algorithms as Pentax.
Nikon seems to use a little weird choice (double size of Sony).
So we have anything between one Byte and two Bytes per pixel.

And all four produce very similar results (for the minority who care about results).

[Na Horuk]

And all four produce very similar results (for the minority who care about results).

And I think this is exactly the point. Differences at this point are pretty much negligible. You get greater differences if you use good settings/techniques (vs bad or automatic), if you use better PP programs and techniques. Sure, there are some differences between different sensors, but these are getting more and more narrow; until you start comparing them to Foveon or something exotic like that.
If one looks at a raw file with zero modifications, no NR, no sharpening, no added contrast (all of which get applied automatically by most PP).. then the file looks boring. It doesn't look like what the photographer saw. It doesn't look like what the audience wants to see.

I remember reading an article that claimed that it is very smart to use an algorithm that applies NR and compresses data at the same time, as you lost practically no detail, but the image looks much better, and is far easier to store. This article even argued to use a lower bit rate, since most extra bits only contain noise. Only "expand" to 16bit once you are on computer and starting to do PP. I don't know how correct this article is, but it certainly raised some good points

[Barry Pearson]

Ah ah I know, but I honestly simply haven't reverse engineered a DNG, nor JPEG and I don't work for Ricoh, so I can be sure if it is lossless or not.

Fair comment.

I have published lots about the contents of DNG files. Both from reading the specification, and from my own examination of the contents using hex editors, etc:

(Barry Pearson's) DNG - Digital Negative format