Slashdot Mirror
JPG Compression - The Bandwidth Saver
Mr.Tweak writes "TweakTown has posted an article entitled "JPG Compression - The Bandwidth Saver". An article for webmasters and site owners showing how they can significantly reduce the amount of bandwidth they use by compressing JPG images, one of the most common formats for web images. If you own a website and don't yet have knowledge in the field of JPG compression, you should find this very interesting indeed - Save money on bandwidth and please viewers at the same time with quicker loading webpages. They also talk briefly at JPEG2000."
The article didn't discuss the merits of other formats vs. JPEG enough. Namely, non-photograhic images, with few colors and mostly large solid areas, such as icons, bullets, line drawings, diagrams, charts, etc. are not optimal for JPEG.
In such cases, GIF and PNG will yield much better compression than JPEG, and also look nicer, since they're lossless. Compressing such images with JPEG will give you ugly "ringing" artifacts, since the lines are essentially infinite-frequency "spikes" which you can't capture completely.
There's 10 types of people in this world, those who understand binary and those who don't.
I got trapped into clicking the link to the article, thinking it could be something useful about jpeg compression and how it works. The story certainly implied that. How disappointed I was, and I agree with most people that this is one of the dumbest stories ever. This kind of story belongs in a site for amateur wanna-be web designers, not geeks. Anyway, since I've gotten all caught up in this, and you're here too, maybe we can have a useful discussion on How JPEG Compression Works.
:), only the associated values need to be stored. I believe the first image deals with how much darker the left half is than the right half of the image, the second does the same thing horizontally, the third deals with how much darker the leftmost quarter of the image and the third quarter of the image are darker than the other two, and so on...
I am by no means an expert, and I believe this is a gross simplification of the process, but here is what I think happens. The jpg alg breaks the image apart into 8 pixel by 8 pixel subimages. (Don't ask how it handles pictures that are not n*8 x m*8 in size). Then it treats each of those images with a process very similar to principal component analysis, where a set of representative images are given associated multipliers of how much of that image to add into the reconstructed original image the user is trying to get. These representative images are ordered from least to most detailed, and since they are known to both the compressor and the uncompressor (depressor?
So how does one adjust image quality / compression? Well every possible 8 x 8 picture can be represented with 64 of these representative images. However, since the 64th deals with *really* minute details, then you can get a decent reconstruction using just 63. It all depends on the image you are trying to compress, but can probably get away with even just the first 20 of the basis images. Oh, for the record, I'm talking about grayscale here. I think you'd need to ramp things up by a factor of 3 to do rbg.
If someone wants to fill in any gaps or factual inaccuracies, certainly do so.
Although jpg compression is definitely helpful, the article forgets to mention that two image formats are supported by all browsers. GIF being the second.
In addition, 4.0 and newer browsers support Portable Network Graphics (PNG).
GIFs should be used for vector based graphics
No they shouldn't. Use PNG for still images. Use SWF (now an open format) or MNG (not much browser support yet but works in Mozilla and Konqueror) for animations.
and provides a better overall quality/size advantage when done right.
PNG can be 10% smaller than GIF when crushed properly.
Will I retire or break 10K?
At my last job, we wondered why our carefully tuned images looked like shit on AOL. We found they were recompressing our jpegs to make them much smaller (and thus lower quality). So we now send AOL really high quality jpegs so that our images don't get trashed as badly by AOL.
Plato seems wrong to me today
I'd like to see some kind of lossy PNG
That's JNG (JPEG Network Graphics) which is JPEG wrapped in PNG-style chunks along with an optional alpha (transparency) channel. Mozilla will display them and IrfanView will process them.
The JNG spec is available somewhere on the PNG web site, http://www.libpng.org/pub/png