Interesting discussion on new Apple codec:
https://news.ycombinator.com/item?id=15342332
Interesting discussion on new Apple codec:
https://news.ycombinator.com/item?id=15342332
Actually, it is not an Apple codec. Yes, it's also driven by Apple, along with Nokia IIRC, most notably Dave Singer (whom I also know in person), though HEIF is not "a new codec". It is a file format around HEVC that adds metadata on top of the HEVC codestream. Thus, nothing really new here - and essentially the features the JPEG 2000 fileformat (notably JPX) could also do. Compression performance: Quite good, of course, also for stills. Also quite complex, of course.
Perhaps I have misunderstood something, but ...
... for me, HEIF/HEVC/BPG/etc. is a big disappointment in quality. Many of the images in the test benchmark I use for image compression are consistently destroyed by these codecs. Particularly bad is the loss of details in red areas -- they are often completely removed and significantly blurred. At qualities around 95 and above, i.e., "camera qualities" HEIF/HEVC seems a worse solution than even the old jpeg.
PIK and guetzli/PackJPG are the only system that competes with JPEG at high qualities -- the modern video codecs don't seem to be doing it.
I really do wonder if Apple has some magical solution for keeping all details with HEIF or if they stopped caring about image quality...
khavish (28th September 2017)
Sorry, but compression performance means "rate distortion performance". At which rate, do you believe, does JPEG provide an advantage over HEVC? Can you please provide examples? In my experience, it is also a matter of people getting used to JPEG defects, for example oversharping defects at edges.
My interest in lossy image compression is primarily in high quality (although transferring now to lower qualities, too). I see that YUV420 works with only about 5 % of the images that I play with, and can spoil some visual features in a rough way. So, I do comparisons with YUV444.
I don't care about PSNR/SSIM vs file size. I want to compress a photograph so that none of its features gets corrupted.
With modern video codecs that seems to be very difficult -- at least with the opensourced software codecs there tends to be changes in details and noise characteristics (some gray wall with surface texture may completely lose its texture, or particularly red surfaces may lose details in white-black direction in them).
When I change the settings to a quality that keeps them, the file size is more than what I get with something like guetzli at quality 95.
With my test corpus at https://drive.google.com/drive/folde...Fk?usp=sharing one can try bench.png or red-room.png. Typically video codecs remove the surface details of the bench, and de-decorate the wallpaper on the right side of red-room.png
Actually, there are some local supports from Apple when it comes to HEIF.
To save photos at a later time as JPG format, you can:
go for"Settings">"Camera">"Formats" and choose"Most Compatible"
To change HEIF format to JPG when tranfer to Mac or PC, you can:
go to"Settings">"Photos">Transfer to Mac or PC and choose"Automatic".
However, you can only take the aid of workable heic to jpg converter to convert the existing HEIC images to more acceptable format.
That's why we make subjective tests. I would believe WG11 also makes subjective tests... At the end, it depends on the test protocol, of course. If you use a single-stimulus test with a 5-point rank scale, the quality might be excellent, but is probably not visually transparent (quality vs. fidelity). There are also tests for subjective lossless coding (AIC-2 does that), but results again depend very much on the test conditions and on monitor calibration. Hence, what might be transparent for you might not be transparent for somebody else - and vice versa.