Bigger Than Big, Taller Than Tall

October 7, 2005 6:24 PM

PhotoshopFilter: How does one add resolution to an image? My current image size is 3'w x 2'h at 100dpi. The finished image needs to be roughly 10' x 7.5' (possibly bigger) at 100dpi. If I just scale it up or add dpi, I'll have noticable pixelization.

I believe there's a way to resample or reinterpolate the image with either Photoshop or a special parsing program that somehow adds resolution to an existing image so you can blow up small images to a very large size--I think billboards companies use something like this.

posted by fandango_matt to computers & internet (31 comments total)

Well...not really. There's no magic TV-detective-show type technology to make a blurry image clear, and what you are basically dealing with is not having the image information at all to start with, same as they are.

Billboard people -start with- very large images, and they are pixelated (well, different, but you get what I mean) when you get up close to them.
posted by Kickstart70 at 6:42 PM on October 7, 2005

You can use a sharpen filter to make a blurry image (that's basically what you have when you try to enlarge it) a little crisper. I like to use a highpass filter - create a duplicate layer, apply the highpass filter to it (you'll have to experiment as to what looks good as far as pixel width) and then set the blending properties of the highpass layer to "overlay." It pulls out bright spots of the underlying image to give you more definition.
posted by chocolatepeanutbuttercup at 6:51 PM on October 7, 2005

You want to scale it up, but make sure you use bicubic filtering. This will mean no pixelation, but it will be blurry - you cannot add information this way.
posted by polyglot at 6:51 PM on October 7, 2005

Genuine Fractals might be what you're looking for.
posted by whatisish at 6:52 PM on October 7, 2005

There are several expensive upsizing applications out there, but they aren't magic and can only do so much. I don't think you're going to have much luck scaling something that big, but it's hard to tell without seeing the original.
posted by glenwood at 6:53 PM on October 7, 2005

There are various scaling algorithms that can (sometimes) do a better job than the bicubic resizer in Photoshop, like Genuine Fractals, stair interpolation, etc. Here's a good comparison of them. But in your case, I wouldn't expect miracles. You can't get something from nothing. You ever seen a billboard close up? They're super-blurry, and the halftone dots are 1/8"-1/4" big.

You may be in luck if image you're trying to scale a very basic line image, in which you can trace it to a vector format with the tracers built in to vector drawing tools like Illustrator or a program like potrace. Once it's vectors, you can make it any resolution you want and it'll be pristine.
posted by zsazsa at 6:54 PM on October 7, 2005

Other answers are correct, if the image data isn't there to begin with, the best Photoshop can do is "interpolate" or essentially guess what it thinks should be there based on surrounding pixels when you ask for more resolution. The more you ask for, the more it has to invent, and the overall image quality will drop accordingly.

If you're unsatisfied with Photoshop's upsampling capability, you can try to use a 3rd-party plug-in or application. There are some programs that use different methods for upscaling (such as fractals) to generate arguably better results than Photoshop's built in algorithms. Try Google for those.
posted by robbie01 at 6:55 PM on October 7, 2005

I should add I'm a graphic designer/art director, using Photoshop CS (OSX) on a daily basis, and I speak advanced/fluent Photoshopese. The graphic is for a tradeshow booth, and the output will be at 100 dpi with some sort of inkjet/film technology. Please feel free to get technical with me!
posted by fandango_matt at 6:58 PM on October 7, 2005

As others have mention, you can't truly increase the resolution of an image without some source of additional information. However, if you have a sequence of related low-resolution images of the same subject (such as a series of video frames) with sub-pixel motion, you can perform motion compensation to align the images and then sub-pixel sampling to construct an image with a higher resolution than any of the individual low-resolution images. This is rather common in the fields of astronomy and video analysis.
posted by RichardP at 7:15 PM on October 7, 2005

Up-rezzing usually means first scaling up, then gently gaussian-blurring out any artifacts, then equally gently adding unsharp mask or the like. Although I will be trying chocolatepeanutbuttercup's technique soon too.
posted by zadcat at 7:18 PM on October 7, 2005

With the dimensions you provided you are scaling your image approximately 375%.

I just did a quick test using those numbers with Genuine Fractals 30 trial version and I'm not too impressed. The standard bicubic interpolation is really about the same.

Bring the image into Illustrator 2 and Live Trace it. Turn your pixels into vectors.
posted by jeremias at 7:35 PM on October 7, 2005

You can get very good results by incrementally scaling the image up in steps by about 10% each time until it's the size you need.
posted by Caviar at 7:35 PM on October 7, 2005

Caviar, what program are you using? If it were that simple, don't you think the program authors would simply break down resizing operation in to smaller pieces like you suggest?
posted by odinsdream at 7:48 PM on October 7, 2005

The results you get by upsizing 10% at a time will be "very good" by comparison to doing it all in one fell swoop, but they'll still suck.

There is no way to do what you want. You might be able to get away with some kind of art filter that turns your uprezzed, blurry image into a collage of faux brushstrokes, however. (Try the photoshop Sumi-e filter.) Or make a photomosaic. Either will give you lots of "detail," by which I really mean visual interest, so the image won't look like total crap close up. But you'll have to step back to see what it actually is supposed to be.
posted by kindall at 8:20 PM on October 7, 2005

Using Photoshop. I originally got this tip from one of Scott Kelby's books, and in checking his new CS2 book, he points out that they did actually add this to the image size dialogue box.

I haven't tried it, but according to him, you can get very good results if you just type in the new size and resolution in the image size box, and use "Bicubic Sharper" in the resample image selector.
posted by Caviar at 8:23 PM on October 7, 2005

Er, dialog box.
posted by Caviar at 8:23 PM on October 7, 2005

Okay, well, now I've tried it, and that's simply amazing. Do that.
posted by Caviar at 8:30 PM on October 7, 2005

Illustrator CS2 has the Live Trace feature that does a pretty good job of converting bitmap to vector. As said above, once it's vectorized, you can do whatever you want.
posted by Hildago at 8:56 PM on October 7, 2005

I'm with most people here -- you can't expect great results from anything that tries to add information that simply isn't there -- but I also wanted to weigh in re: the recommendations about Genuine Fractals. Having worked with digital photo manipulation at a major magazine for about a decade, I'm firmly in the came that GF is amazingly overrated, and doesn't do one whit of a better job than the worst option set in Photoshop. Don't waste your time there.
posted by delfuego at 9:02 PM on October 7, 2005

I'm going to go with Hidalgo: I've been faced with this same problem many dozens of times, when preparing art for output at very large sizes (try 30 x 10 feet). Two hours tracing your art in photoshop - for example, using a wacom tablet and live trace and the magnetic lasso tool - and then compositing your paths & layers and adding color in illustrator - is worth ten hours trying to fake it with unsharp mask and resizing plugins in photoshop.

No matter how expensive or mathematically complicated the resampling technique, you'll probably still end up with junk. You cannot add information that wasn't there to begin with, unless you want to mask the blur with noise detail.
posted by luriete at 9:35 PM on October 7, 2005

I have had pretty good luck with the above incremental method over the years, although it's touch and go sometimes. I figured this out through trial and error over a few years. I usually used the bicubic option. Every few iterations, try a blur and/or sharpen filter. You can sometime use the clone brush on a middle range opacity clean edges or to randomize artifacts. I have not used any current versions of Photoshop, so I am unsure about this added option in Photoshop that Caviar mentions. If you do it yourself you will have more control. I have done this with Paint Shop Pro as well.
posted by Yorrick at 1:36 AM on October 8, 2005

Yeah, I'd second (or third or whatever) what Jeremias, Hildago and Luriete said. The new live trace feature in Illy CS2 is a lifesafer. A client recently asked me to create a 2000mm x 800mm banner for him using a photograph (which measured approx. 100mm x 80mm at 300dpi) as a prominent - read large - element of the design. After a lot of futzing about using bicubic sharper interpolation and a mix of gaussian blur-unsharpen mask-high pass filtering (the last rocks, BTW on digital camera images > print), I decided to give Illy CS2 a go, and I was mightly impressed with the results after playing with the settings a bit. Give it a go, it's pretty sweet...
posted by ninthart at 2:21 AM on October 8, 2005

you can also rescan the image at a higher res and then resize bucubic.
posted by miss tea at 7:43 AM on October 8, 2005

I'm surprised that only one person has mentioned halftone dots. At such a large format, I have to assume you're dealing with a printed medium. In the printing process, there are actually several independent numbers inaccurately called “dpi.”

One of these is the resolution of your image, which for the sake of distinction, we can call “pixels per inch [ppi.]” Another is the resolution of the printer’s halftone screen, which we can call halftone “lines per inch [lpi]” or “screen frequency.”

I think your estimate of your resolution requirements is too much.

The long and short of it is: 10’x7.5’ is big. You want to upscale your image by over 300% and you’re not going to be able to “add” resolution, no matter what. Fortunately, perceiving such a large image requires a pretty great distance between the subject and observer. (You mention billboard companies; remember how far away the billboards are from the highway?)

That means you can get away with a much smaller screen frequency. For a point of reference, newspapers print images at around 85lpi. Magazines print images at around 150-200lpi. According to my Hayden Shop Manual, the industry standard [in 2000] was to prepare images with a ppi of double the target lpi.

Now, if you scale your image up to 11’x7.5’, without resampling, you'll have 26.67 ppi. You just might be able to get away with printing at 10-15 lpi. [When you're 25 feet away, a .10” dot is pretty small.]

Have you talked to a printer? You’d be well off to talk to one who specializes in large-format printing. I have an uncle who used to run a large-format printing business. He’s out of the business now, but if you can’t find anyone in your area, I may be able to put you in touch with him.

Also, chocolatepeanutbuttercup’s method of sharpening with a high-pass filter is pretty effective. I've found that with the digital photos I get from my Canon, it often works even better to apply this technique to individual channels. I generally do red first with overlay blending at 80-100% opacity, then green with soft light blending at 50-60% opacity. Most times I don't bother with blue since that channel is always the noisiest.
posted by ijoshua at 8:27 AM on October 8, 2005

D’oh! I missed your followup. What I wrote may not be very much help.
posted by ijoshua at 8:42 AM on October 8, 2005

I just had this idea. If your ouput must be 100 lpi, you might try running the Color Halftone filter on the scaled photo before putting it in the final layout for printing. This will give you a stylized effect with more aesthetically pleasing pixellation that should still look great from a distance.
posted by ijoshua at 8:56 AM on October 8, 2005

Have you considered rasterbating?
posted by ulotrichous at 9:43 AM on October 8, 2005

if you don't have cs2, then do multiple applications of slight gaussian blur followed by slight unsharp mask. you'll have to eyeball the results to get the desired effect. 375% is a lot, but I've had decent results at a 200% increase with that technique. blur, sharpen, blur sharpen...
posted by shmegegge at 12:38 PM on October 8, 2005

i work for a sign company, and we make large format trade show graphics allllllll the time. basically, what a lot of people have said here is true: you can't ADD information to a small image. but you do have some options:

first, consider viewing distance. if people are going to be a ways back from this image, then you can get away with more pixellization than you would if people were up close looking at it from a couple of feet away. a lot of trade show booths are set up so that the customer is never going to be closer than 10 feet away, and you can use this to your advantage.

i've used Genuine Fractals occasionally to get a little extra boost to the resolution. they're website claims best results at between %500-%800 enlargement from the original file. it tends to work best on photographic images, and not so well on anything with crisp edges or a lot of vector-looking content, including text. it can help, but it's not going to get you a perfectly sharp image. what it does is help hide the inevitable pixellization a little better than a straight resizing in photoshop will.

another thing that works to your advantage is that a lot of wide-format inkjet printers use what's known as stochastic image processing, meaning instead of a regularly spaced halftone pattern, it's going to print a somewhat more random array of colored dots, which helps minimize the appearance of pixellization as well.

good luck...
posted by cathodeheart at 1:44 PM on October 8, 2005

one last suggstion, though this is off the top of my head and I'm not sure how well it would work.

have you considered applying a half-tone pattern to the image to SIMULATE printer half-toning and therby cover up the elargement artifacts? then people who see the image up close will assume that it's a normal large print pattern. you'll want to be careful how large you make the pattern, though. too big and it'll look fake.
posted by shmegegge at 8:30 PM on October 8, 2005

Coincidentally, I just picked up the September/October Digital Photo Pro magazine, and there's an article on exactly this topic, which covers a few techniques.
posted by Caviar at 2:35 PM on October 9, 2005