File talk:JPEG JFIF and 2000 Comparison.png

Compress some text, too

Might be useful to include the text captions in the picture before compression so we can see how artifacty the text gets. If that was already done then perhaps point out the fact? Tempshill (talk) 06:18, 28 March 2008 (UTC)[reply]

Is it really better?

Am I wrong here, or does the standard JPG look better than the JPG2000 in this sample? The standard JPG compression artifacts are organized in such a way that it produces the illusion of a sharper image. The JPG2000 artifacts just make the picture look blurry, for the same given size. The only advantage I see, is that the standard JPG has an alpha channel, making it more like a lossy compressed alternative to PNG. Am I wrong? --76.203.214.40 (talk) 13:48, 3 July 2009 (UTC)[reply]

I think there's a problem with this image I keep hearing jpeg2000 is better than Jesus, but olde Jpeg looks better on that image.Ren ✉ 17:13, 4 November 2009 (UTC)[reply]

I agree with you that there seems to be something wrong with the example image. Still I'm quite sure the JPEG and JPEG2000 images are not swapped because the middle picture shows very typical JPEG compression artifacts. Someone should try to make a new example image. -- 84.188.255.45 (talk) 11:22, 8 December 2009 (UTC)[reply]

This is definitely a very bad image. A new one should be made, and it should be mentioned which encoder is used (for JPEG and for JPEG 2000), as that can make a big difference. The JPEG 2000 here looks like it was encoded by a drunk old man with an abacus. 128.248.9.61 (talk) 19:18, 25 April 2013 (UTC)[reply]

Redoing an experiment because the outcome doesn't fit your expectations is not scientific. From my experience, JPEG2000 consistently delivers inferior quality, at least at bitrates where you get negligible artifacts with JPEG. And you don’t even have to use JPEG to its full potential for that. Using a better encoder like mozjpeg, using arithmetic coding (which has been in JPEG forever and meanwhile is even patent-free), and better decoding technology or post-processing that avoids blocking artifacts (e.g. DOI:10.1023/A:1008167430544), JPEG quality can be improved much further. JPEG2000 might have the edge at lowest quality settings, but that's not what you'd want to use in most cases 'cause it looks horrible, JPEG2000 or not. So, yeah: good old JPEG is the better choice in most cases, not even talking about the benefits of going with the prevalent format.--Kulandru mor (talk) 11:49, 2 April 2019 (UTC)[reply]

How was the JPEG 2000 Version generated?

I think it should be mentioned by the copyright holder how (i.e., what tool/process they used) they created the JPEG 2000 version of the file. He does a fine job documenting how the other parts were created. — Preceding unsigned comment added by Dan Aquinas (talk • contribs) 22:09, 11 March 2014 (UTC)[reply]

Reverted change by Nition1

I reverted Nition1's new image because the JPEG image looked ridiculously awful considering its file size. I suspect the encoder (Photoshop) was inserting a large XML data block and/or thumbnail image.

In contrast, when I tried to reproduce the original image I got reasonably close results. The JPEG image (IJG quality 30, non-progressive, 2×2 chroma subsampling) looked identical but was actually smaller (10.3 KiB), so I suspect the original encoder didn't use optimal Huffman codes or added some unnecessary metadata. The JPEG 2000 version (which I saved at a target size of 11469 bytes ≈ 11.2 KiB even though the JPEG image was smaller) looked worse than the original (less detail), but this could be chalked up to using an inferior encoder. I used the LuraWave demo encoder that ships with IrfanView; the original encoder is not specified. -- BenRG (talk) 22:32, 3 January 2015 (UTC)[reply]

I was able to reproduce the original JPEG 2000 image exactly using JasPer (GeoJasper 1.3.1 incorporating libjasper 1.701.0, official Windows build, with -O rate=0.03). The file size is 11499 bytes, which, again, is almost 10% larger than the exactly matching JPEG image produced by cjpeg -opt -q 30 (10494 bytes).

I also tried Kakadu (kdu_compress v7.4, official Windows build, with -rate 0.65) and got a totally different-looking result. JasPer seems to preserve prominent edges better at the expense of subtler details, while Kakadu blurs everything more uniformly. -- BenRG (talk) 00:22, 4 January 2015 (UTC)[reply]

You're absolutely right and I apologise for my misguided edit. I always thought it strange that the JPEG example looked better than the JPEG2000 (as others obviously have given the discussion above), and my Photoshop experiment seemed to confirm it. However, the error I made was saving the JPEG via "Save As..." in Photoshop instead of "Save for Web & Devices...", where The former obviously adds some cruft to JPEG but not to JPEG2000. Testing it again now with save for web, I get essentially the same results as the original upload. -- Nition1 (talk) 08:05, 12 January 2015 (UTC)[reply]