HD For Indies

Raw notes follow:

JPEG2000

latest ISO image compression standard

ISO 15444

published jointly with ITU

published in multiple parts

part one is royalty and fee free

JPEG2000 has 12 parts, and not all parts are complete

part 1 is core coding system, includes minimal file format compatibility, everyting you need to encode/decode JP2

part 2 contains various value added stuff, enhanced performance and niche markets- hyperspectral, multispectral, space stuff, etc.

part 3 is motion JPEG 2000 that includes timing and audio stuff

DCI is NOT using JPEG2000 motion imaging stuff, they are using MXF

part 4 - is conformance stuff

part 5 -references software in

part 6- for document imaging - compound text/graphic images

part 8 - watermarking and encryption for security, DCI not using this

part 9 - client/server protocol for imaging huge images using scalar/layered/tiled stuff I'd guess

part 10 - 3D and floating point data for medical and scientific - CAT & MRI stuff

part 11- wireless - help for protection from image errors on noisy comm channels

part 12- media format for common architecture for MPEG-4 & JPEG2000

part 13 - reference decoder

JPEG2000 FEATURES:

very scalable code stream (extract lots of image products from a single stream)
progressive transmission - good for transmit over slow links, can organize how they send it - progressive download, 4 different kinds

-lossless and lossy (available naturally by progression)
-handles binary as well as continuous tone
-random code-stream access
-can do region of interest to vary quality in one region as opposed to another can rob bits from other parts to enhance another
-and of course, very high quality

take your high res source, compress to 4K JPEG2000, can extract a 4K or 2K from that without downsampling - just pull the bits we need for 2K, decompress and display. Can do a 1K extraction for CRT's and stuff. Not great for cinematic display, useful for workflow
-can extract a piece of an image without having to read the whole image and crop it, can find the piece you want and only decompress that

progression by resolution - 64x27 to 128x54 to 256x108 to 512x215 up to 4096x1714 (was his scope source)

decode a little bit of data and you get progressively more info

can do progressive downloads for enhanced quality - just like progressive JPEG on the web - multiple passes makes it better

he shows a zoomed in stuff - looks quite quite good at 85 Mbit. 250 is the standard

can see some small changes in grain structure between 24:1 and source

can load by raster lines is another option.

JPEG 2000 ALGORITHM - how it works

Decorrelating Color transform

typically RGB, but X'Y'Z' for digital cinema possible to

no subsampling

applied indepently to each color pixel

most energy is in resulting Y component - the other two factors are highly compressible

this itself improves compression efficiency
2:1 rate improvement for same quality

there's a lot of correlation and redundancy between Y, Cz and Cx...by stripping out some redundancy, it's still mathematically possible to get a perfect recreation of the source, but you've made compression much much easier

then a wavelet transform - filtering and downsampling

by downsampling part of it, and doing a high pass, you create a way to still make a mathematically perfect source, but still takes up even less dataspace

3 kinds of high frequency detail that get recorded on top of the half res first part.

bit plane coding - high and low quality extractable from save file - you start slicing off bit planes - take the most significant bit and compress it and put it in the file, then the next significant bit and compress that, etc. etc. etc. Not just resolution sampled, but bit order sampled

codeblocks and precincts - all the data is divided into....shit I don't understand. It's all organized into packets, so each chunk of the image is written into it's own piece. NOT a row then a row then a row, it's like tiles. You can access just the tiles you need.

SO..it's tiled geographically - you can access just the piece of the image you want. And it's tiled in terms of image quality - you can dial in just the resolution you need

Q&A:
-------

Q: MPEG can have good and bad encoders - is JPEG2000 like that?

A: encoder quality can vary, but it's not nearly as critical as it is in MPEG. It's easier to build a good JPEG2000 encoder than a good MPEG encoder. There will be better and worse

Q: How granular can you get on your blocks (physical size):

A: user selectable 4x4 up to 64x64 as long as the footprint isn't bigger than the 4K wavelet coefficient. For digital cinema, it's 32x32 pixels.

Q: What are some constraints of j2k for d-cinema usage?

precincts are locked down to 256x256 (last two spatial geometrical regions shown) a bunch of specialty modes in j2k encoder, for cinema you're locked to default mode to be most compatible for interchange, bitrate is constrained to 250 mbits/sec, image must be compressed as a single tile (can tile image before all this), it's forbidden in DCI spec

# of wavelet transforms is flexible, 0-5 for 2K, 0-6 in 4K, progression order is fixed,

Q: some aspects are variable - would the wavelet transform itself be variable?

A: No. The 9/7 is used for the DCI stuff. "The quality of the 2K extract from the 4K is quite good."

Q: It's an intraframe compression technique - are there any problems with motion, frame to frame, that only show in motion?

A: At the bitrate in use, it's quite OK. Many algorithms were tested, JPEG 2000 was best picked on 50 foot screen. For visual quality and quantitative quality, intraframe JPEG2000 was up there with best MPEG. "You don't buy much by doing the inter-frame stuff."

Q: how does metadata survive encoding
A: metadata is separate - color space isn't even known to JPEG2K, it doesn't care. DCI will be using MXF to carry the metadata

Q: guy from Kodak, about 250 Mbit/sec spec. Backwards compatibility is important, but we've built a hamstring w/250 Mbit. Aren't we sticking ourselves with this bitrate?

A: many arguments over this, "a lot of people were involved in the testing, prepared material, testing done side by side with butterflies with original and compressed content" the folks looking were satisfied it would do the job.

comentary based on conversations: the 250 Mbit is a contentious issue - evaluations were down for 2K @ 24fps at 250 Mbit...not 2K @ 48fps, not 4K @ 48fps. 250 is what's doable with current technology...there's arguments as to the efficacy of 250 for 4K over time. May have hosed ourselves based on what's doable now. A 2.0 spec in future? Have to have multiple deliverables if did that, against the point