I’ve been doing a lot of Photoshop work in the past year. The images I’ve been collecting will mostly be used for PowerPoint presentations, but someday I may want to output them in hard-copy form as a laser or photo printout, or even as part of a professionally offset-printed document.
When I’m through fiddling with an illustration and I’m ready to save it for good as a jpeg, Photoshop asks me to assign it a degree of quality on a scale of 1 to 10. I know that a low number means a smaller-sized file, but what exactly is being lost?
(I’ve always had a theory, but I’ve never read anything to confirm it. I always assumed that on a lower setting, fine differences in shades of a color get smoothed out. For example, all near-yellowish pixels would be assigned one generic yellow color. The lower the setting, the wider the shade-difference net. Is that right?)
And how low a quality is too low, and how high is just overkill, given my above-described needs?
Not exactly – what you’re describing is closer to the way .gif compression works.
.jpg compression is a bit more complicated. Lower quality jpgs are less fussy about averaging colours, but you’re not really reducing the number of colours directly. You’re assigning the same colour to larger areas by being less fussy about the difference in shading. Lower quality jpgs have big ol’ blocky bits where detail has been sacrificed.
Look at it this way – if you have a black and white line drawing with many fine, detailed curves, are you going to save it as a gif or a jpeg? A “lowest quality” .gif (with 2 or three colours) with contain close to the same information as an uncompressed bitmap, because all of the pixels are represented. A “lowest quality” jpeg will try to cut corners by squaring off large areas and assigning one colour to them – okay if you have lots of large areas of solid colour, but no good if you have consistently fine filigree.
Really, though – Don’t “save it for good” as a .jpeg. Even if you are certain you’ll never need to fiddle with layers or anything, you’re throwing away detail. If you’re really stingy with space, save it as a .bmp or a flattened .psd – this is a good idea even if you are going to keep the .psd, which there’s little reason not to. Bitmaps may be useful if you lose fonts or something.
You can see this in “real time” if you choose “Save For Web” instead of “Save As” and jpg.
The “Save For Web” has a preview of the image and every time you change a setting (there’s more settings in there than just quality) you can see what the output will be right there. You can also switch between the “optimized” (your settings) view and the original to compare.
I know this option is available in PS 6 and above but I can’t recall if it was in 5 or not.
:smack: Upon further inspection I realized this method will by default save your pics at 72 DPI. So…don’t use this method to save, but you can use it to at least preview what might get lost in your images.
Regardless of what graphics program you may use, always, always, work in the native format of that program. At the same time, use an image quality level commensurate with the quality level of the highest publication level you might use. So for example, if you plan to publish a full-color hard copy, consider working at 300 DPI, or even 600 DPI.
Then when it comes time to produce a publication copy of the image, be it for print or the web, make your publication copy from your native master. Your 300 DPI native master can then produce a 72 DPI .jpg for the web, or even a 300 DPI .jpg for print.
I think more specifically what you are throwing away is the higher spatial frequency components of the chromaticity (the ratio of primary color contents), especially in the yellow-bloe dimension and somewhat less in the red-green dimension.
Imagine you start with a photo in which each pixel has R, G and B values. You could call the average the brightness (or call the G the brightness like color television does). You are going to keep all the resolution of the brightness.
But suppose after you create a single channel picture with brightness, you then make an image of red-green chromaticity, and another picture of blue-yellow chromaticity. Then you Fourier transform these two, and you discard all the amplitudes for spatial frequencies above two limits (higher frequency for RG and lower for YB), and save these two smaller pictures.
When reconstructing, you fluff up the two smaller pictures by adding zeroes all around the central regions, then you inverse Fourier transform them. Finally you add all these together again.
JPEG doesn’t do exactly this, but it does something kind of similar (if cheezier and based on square subpanels). The thing it does takes advantage of the same fact - that human vision can’t see the chromaticities with much sharpness. This is the reason it’s so hard to read yellow letters on a white background, or blue ones on a black background.
You can convert a picture to JPEG and then back into RGB, and do a principle component analysis on the resulting three chanels, and make images of the three eigenvectors. The first image generally looks like a good BW version of the original, and the second often looks like it encodes how warm the colors are, say. The third one will be an ugly mishmash of square panels with broad shading bands in it. That’s because JPEG carries less information than you need to make the color image, and the method has a way of making up information that you don’t much notice anyway - sort of sweeping the information loss under the carpet, from the point of view of your visual system. PCA will pull that information together into one place where you can see it clearly, hence the patchwork 3rd component.
I personally archive my stuff as 600dpi PNG files, and everything I’ve done has printed beautifully. Stuff like book covers, art prints, flyers, posters, etc.
You want to heed this poster’s advice though, and be saving your files you may want to print as PSDs (or PNGs or possibly TIFs). Save things at the highest resolution you have space for. Basically you need a lossless compression/image format. The biggest problem with JPG compression is that everytime you resave it, you lose quality. If you save it a low quality, not only will it not print well when you require a high resolution, but if you have to edit it and resave it, it will look even worse.
JPEG is a “lossy” compression algorithm (Actually, there is a lossless version of JPEG, but no one seems to use it.)
When you take a small JPEG image and expand it, it looks very fuzzy and “blocky”. As previous posters have mentioned, a lot of detail has been dumped in the name of small file size. A lossless algorithm, such as PNG (which, unfortunately, is not supported by all the big browsers if this picture is destined for the web) will have the same level of detail no matter the viewing size.
Theoretical answers are all well and good, but I think what will help the OP most is actual experience. To see the effect most strongly, create a simple black-and-white line-art image. Make sure that most of the lines are not just vertical or horizontal. Now, save that image as a jpeg at low quality. Close it, and re-open it. You will see the difference.