Quo Vaidis JPEG ... the child has a name, and a number.

[thorfdbg]

Hi folks,

it's probably more than time for an update on this, and sorry for being quiet for such a long time. To remind you, all this is on future JPEG directions and our future standards. In case you did not follow, our latest baby will be a backwards compatible JPEG extension that includes all the features JPEG should have had originally, but did not. Like HDR coding, lossless coding, alpha channels and a lot of features I forgot. All that, decodable (though probably not lossless, or at HDR quality) with your legacy applications. As your browser.

---

So this time, we finally have a name, and also an ISO number. The name is JPEG XT (for ExTended). Looks like we skipped JPEG XS by now. The ISO number is 18477-x, where we have at least the following x's:

---

Part 1: Defines the baseline, and is more or less 10918 (old JPEG) plus color transform and upsampling (JFIF) minus subsampling factors nobody used and supported, minus arithmetic,hierarchical and old predictive lossless. We're going to have a committee draft version of this soon.

---

Part 2: will cover HDR coding, i.e. more than 8 bits per pixel, likely floating point, and probably some other stuff. Lossless will probably be separated out into another part. Lossless technology is there, HDR technology is there, though evaluation is still running of what will be included.

---

Part 4: will be conformance testing. Part 3 we did not forget, just a placeholder.

---

Part 5: will be the reference software.

---

Software updates will follow as soon as I can provide them. We're right now still fighting with finding a suitable license, something that is more liberal than GPL, but will also enforce that the stuff is only used in JPEG XT conforming software. I'd guess we like to avoid people drilling up the stuff and calling it JPEG for reasons... oh well. More to follow - live from Incheon.

[Piotr Tarsa]

What about a reference hardware design for the most time consuming algorithms? Maybe sources in Verilog, VHDL or something like that?

[caveman]

Thanks for sharing!

[thorfdbg]

What about a reference hardware design for the most time consuming algorithms? Maybe sources in Verilog, VHDL or something like that?

To be honest, given my experience with JPEG 2000 part 7 (which was exactly that), I don't think that this is going to happen. Hardware designs are higher investments for companies, and they typically do not want to share these.

---

However, please remember that the actual coding engine (entropy coding) is just old JPEG, so you can basically get away (at least in some profiles) with taking two JPEG chips and add some postprocessing.