jpg output: What does subsampling, DCT, ... do?

[obelisk]

Sorry don't even know how to search for this info. The jpg output page,
what do the items below mean, what do they do, what effects on quality/size etc?

progressive
DCT
smoothinng factor
subsampling

[eL_PuSHeR]

Check the Wikipedia.

Progressive, makes the image loading on a line-per-line basis (interlaced?) to preview it faster (for browsers, etc)
DCT is Discrete Cosine Transformation + Smoothing factor + subsampling are settings to fine-tune JPEG compression. I can't help you further, sorry.

[Drahken]

DCT method:
You'd have to be a mathemetician or something to fully understand it, but essentially it's the compression formula. It's self- explained in the export dialog (not sure about the regular save dialog). Fast is fast but results in a larger file (not much larger though). Slow is in the middle, and float gets the best compression but is the slowest.

Smoothing factor:
Just what it says, it smooths out the image. Zoom to 100% on the preview on the export dialog & try changing it between 1 & 10, you'll see the diff if you look closely. This can marginally improve the visual quality of the image, but it's main purpose is to reduce filesize.

Progressive:
A normal (non-progressive) jpg will load line by line, at full quality, from the top down. A progressive one will load the whole image at a low quality, then gradually increase the quality as the image loads. This aspect of it is purely for use in photos you want on the web. However, there is a secondary benefit to this: A progressive jpg will have a noticably smaller filesize than a regular one (at least on most images). This is in direct contrast to the similar "interlaced" setting for pngs & gifs, which actually INcrease the size of the png/gif.

Subsampling:
Have you ever noticed how a jpg image will usually have washed out colors (especially reds)? That's due to subsampling. The subsampling setting will greatly impact both the size & quality of your image. "Default" is the lowest quality/filesize, the 1x1,1x1,1x1 option is the best. If you use the high quality setting, you can get a jpg that looks almost as good as a lossless image, but at a fraction of the filesize.
(It's a myth that jpgs have to look like crap due to being jpgs, it's just that the vast majority of jpgs one encounters were made using the crappiest subsampling setting.)

[Janis]

Thank you for the explanations! I was wandering about these same options today (and also about the Huffman table's optimization). I suggest these are added to XnView's User guide (Info -> Help).

I've been using the progressiveness purely based on the resulting file size. In addition, if you choose the image to progress in 3, 4, or 5 passes (an option that Adobe Photoshop offers), it affects the file size too. The smallest file size is usually with 4 passes for me. :)

[XnTriq]

• Differing display of JPEG images
• What exactly does "subsampling factor" do?
• Advanced JPEG I/O