Re: JPEG 2000 (was: Multiple image files?)

On Tuesday, January 14, 2014 at 5:27 PM, Frédéric Kayser wrote:
> Hi,
> Setting quality to 50 in Photoshop activated 4:2:0 chroma sub-sampling, but you didn't use sub-sampling for the JPEG2000 file (-s 2,2).

I didn’t know that! I used a 50 quality setting in “Save for Web…” mostly because it seemed like a nice, neutral number… I had no idea that it was the border between such a big change in encoding methods.

As for the specific OpenJPEG option that you mention (-s)… I tried it, and it seems to do nothing when you specify the output quality in dB (with -q). When using -q, the indexes for the chroma-subsampled and not-chroma-subsampled jp2s are identical, and ImageMagick’s compare tool thinks that the images themselves are identical, too.

When I try `-s 2,2` with a compression-ratio (`-r 80`) instead of a PSNR quality (`-q 40.9633`), the indexes *do* turn out differently… but not by much… only by a couple of hundred bytes, and the change seems to only affect the highest-res couple of layers. I used the following commands

```
$ image_to_j2k -i USA_Pro_Challenge/uncompressed/full.tif -x ratio_subsampled.idx -o ratio_subsampled.jp2 -p RLCP -n 6 -I -s 2,2 -r 80

[INFO] tile number 1 / 1
[INFO] - tile encoded in 4.070737 s
Generated outfile ratio_subsampled.jp2
Generated index file ratio_subsampled.idx
$ image_to_j2k -i USA_Pro_Challenge/uncompressed/full.tif -x ratio.idx -o ratio.jp2 -p RLCP -n 6 -I -r 80

[INFO] tile number 1 / 1
[INFO] - tile encoded in 4.118631 s
Generated outfile ratio.jp2
Generated index file ratio.idx
[e@Rusty ~/Desktop/jpeg2000 tests/2014-01-12/_reorganized]$

```

to generate these files:

http://ericportis.com/etc/subsampled_chroma/

(specifically, compare this index:
http://ericportis.com/etc/subsampled_chroma/ratio.idx
to this one:
http://ericportis.com/etc/subsampled_chroma/ratio_subsampled.idx)

Any idea why:

1) chroma subsampling is disabled or ineffective on images whose quality, rather than compression ratio, is specified?

2) subsampling the chroma only affects the bytesizes of the highest-res couple of layers? My guess is that the lower resolution approximations contained within the .jp2 are kind of like lossless decompositions of the lossy full-res file -- the subsampling is only happening at full-res, and therefore won’t help the poor efficiency of partial, low-res loads.

Thanks!

—eric