Help SmugMug evaluate a compression technology
Baldy
Registered Users, Super Moderators Posts: 2,853 moderator
We've been meeting with an impressive team of guys who sell "perceptually lossless JPEG compression." My first two thoughts were, "The dgrin powers of perception are more powerful than you're used to," and "Haven't I heard this song before?"
But looking at their samples on 8x10 prints, we couldn't see a difference between the original JPEG and and their ~30% smaller images (in bytes).
We'd like to check them against images you provide. If you have a few images that you think would make good tests, post them here and I'll post their versions so we can compare.
What makes a good test case? Images with hard, contrasy lines on solid colors and images that contain red often show traditional JPEG compression artifacts.
Dark mountains against a blue sky produce ringing artifacts where the mountain and sky meet. But maybe with this new algorithm, the differences will be seen on other types of images.
For some of your images, I'll produce 30x40 prints on both paper and Duraclear and show them around in a double-blind test.
Why would we do this? For one thing, if our display copies are 30% smaller and perceptually the same, our site will seem faster for most people.
Your thoughts.
But looking at their samples on 8x10 prints, we couldn't see a difference between the original JPEG and and their ~30% smaller images (in bytes).
We'd like to check them against images you provide. If you have a few images that you think would make good tests, post them here and I'll post their versions so we can compare.
What makes a good test case? Images with hard, contrasy lines on solid colors and images that contain red often show traditional JPEG compression artifacts.
Dark mountains against a blue sky produce ringing artifacts where the mountain and sky meet. But maybe with this new algorithm, the differences will be seen on other types of images.
For some of your images, I'll produce 30x40 prints on both paper and Duraclear and show them around in a double-blind test.
Why would we do this? For one thing, if our display copies are 30% smaller and perceptually the same, our site will seem faster for most people.
Your thoughts.
0
Comments
If you´re thinking about using the compressed images only for scaled down displays on the site while keeping the originals for printing (which seems wise and inline with your current policy to me) then I guess your blind tests have to check perception on monitors as well as on print.
I know prints will really show the differences but if images pass severe testing for displaying but fail in printing tests, then the use of the compressor might still be a possibility. Display testing will also be a lot cheaper than print testing.
As a testing method, I suggest creating galleries with test images in all smugmug sizes comparing both methods. The friendly smugmug interface is a lot better for viewing and comparing sizes than the dgrin thread.
I also suggest applying their compression on originals of various JPEG compression settings. Some people out there use very high compressions and the added effects can be disastrous.
It´s also important to test the effect of the order resizing/compression. Compressing smugmug resized images may have a different effect than resizing a compressed image (size-wise and quality-wise). You´ll need to switch the images produced by the automatic smugmug resizer to do this in the gallery interface but that´s not a problem as you have the backdoor keys....
I also suggest testing images that customers have already mentioned when referring to compressions settings in Dgrin threads like these:
http://www.dgrin.com/showthread.php?t=85085
http://www.dgrin.com/showthread.php?t=51724
My 2 cents
Robert
However, please do not compress the originals ... and use the originals for prints.
SmugMug is the best! Always improving.
Facebook ♦ Flickr ♦ SmugMug
SmugMug referral coupon code: ix3uDyfBU6xXs
(use this for a discount off your SmugMug subscription)
Display copies, however, have everything to do with compression and load time. So, are you only talking about the possibility of using a new compression engine for the generation of web display copies? If so, then I think you should do as another poster suggested. Find hundreds of troublesome images and put up a gallery where the display copies are made with the new compression engine and let us all look at them. I personally don't see compression artifacts as easily as some, but I do know there are some who see them quite easily.
If you really wanted a torture test, you'd put two copies of the image in the gallery right next to each other, one with the old technology and one with the new and add a caption that lists the filesize.
Homepage • Popular
JFriend's javascript customizations • Secrets for getting fast answers on Dgrin
Always include a link to your site when posting a question
It would have to survive enormous scrutiny to think about using it for originals. JPEG compression levels have survived enormous scrutiny, but we haven't done anything about it even tho it's a very expensive decision. For example, there is no perceptual difference between a JPEG 12 saved in Photoshop with JPEG 9, except the JPEG 12 is 3 or 4 times the bytes. So all the pro print labs ask for JPEG 10, adding a point or two for comfort and safety.
We encourage JPEG 10 too for a variety of reasons, but accept JPEG 12 when we receive them and don't alter them. JPEG 10 benefits us because our storage costs are lower but also because we can process them faster. They benefit the person who's uploading them, because upload times get three times longer with 12 as opposed to 10.
My reason for wanting to make big prints is I want to know if it degrades, and how. With JPEG we know: it gets ringing artifacts in certain images.
We'd also love to know if there is any perceptual degradation on originals. One reason is we receive quite a number of images that are in a color space like CMYK or are larger than 48 megapixels, etc., and for a long time we rejected them, tormenting the majority of people who just want us to do the right thing, which is most often to process them anyway. Or we have to rotate and they've cropped, so lossless rotation is not possible.
But we often don't know what JPEG compression they used, so we have to pick a number. If we pick 10 we expand and degrade their images if they had picked, say, 7 or 8 like cameras use for many sport shooters. But we compress the original if they picked 11 like some fine settings on cameras, or 12 like many save as with Photoshop.
What if in those cases we used this compression technique?
Another reason for looking at originals is today's display copy is not necessarily tomorrow's. When we started, our display copies went to 800 pixels and were compressed more than the 1600-pixel display copies we have now and artifacts are more easily seen with LCD monitors, which weren't in use by high-end people when we started.
I'd also like to start with a smaller sample set because we found before that when the sample set gets large, each image gets less scrutiny and the testers don't spend the time per image to find subtle differences.
-r
This one seems particularly challenged. It looks brilliant in the original RAW, but JPEGs really have a tough time with the blue gradient filtered sky. Chris.