Techniques to dramatically scale an image

Question

I have been given an impossible task:

Take a picture image (jpg) at 1422 x 1067 px with 100 dpi and scale it to 86400 x 29376 px (that's equivalent to 34ft x 100ft or 30.5 meters x 10.36 meters)

Unfortunately there are times when a boss requests the impossible from their employee. This is one of those times.

More specifically it was requested that "there be no pixelation (at a viewing distance of 2 feet (0.6 meters)."

The easy (and right) answer is: It can't be done.

But that doesn't mean you can't make the best of an awful situation. So what I'm looking for are techniques, or software solutions that specialize in this sort of huge scaling.

I have heard of something in the past that blows up and image using circles. Granted, it doesn't look great from close up but from a greater viewing distance it looks more held together (like pointillism).

I don't think there is a clear answer, which is why I was hesitant asking here. So I was wondering if this could be transformed into a wiki, because I'm sure I'm not the first or last to be looking up information regarding this sort of scaling.

Answer 1

As you probably know, the viewing distance of two feet is ludicrous. If people were going to view whatever this is from two feet, it wouldn't need to be 34 feet tall. When people get up close to something that big, they're used to seeing image issues.

From a reasonable distance (20 plus feet?), Scott has the right idea. Depending on the photo, the Illustrator solution is quite elegant. The file size will be manageable and, with a little work, the result can be surprisingly photographic. If you haven't used it much before, spend some time playing with the live trace settings — they can interact in surprising ways.

If you have to suffer through the PS solution (that file is going to kill you!), do yourself a favor and use an action. I've found the best results with increments of no more than 20% (whether reduction or enlargement). For more modest scaling I have actions set up to go either up or down in increments of 5%, 5 times, with the fifth time doing an unsharp mask. The smaller your scaling increments, the better PS is at guessing the new pixels. Since you're going so far up, you could set you action to do 5 or 10% maybe 15 times and then manually do the unsharp so you can more finely control it.

I used to know of a stand alone app that did this sort of thing a decade or so ago (fractal something or other). As I understand it, Adobe bought that technology or something similar and rolled it into Photoshop. If anyone can confirm I'd be interested to know. It certainly does a better job these days than it used to.

Answer 2

I would take a different approach to this, because the task is like wishing for the moon. There's no unicorn filter in Photoshop yet. (And I'm surprised nobody has so far pointed out that Photoshop's pixel limit for a PSD is 30,000 in either dimension, so 86,400 would only be achievable by slicing the image into separate files and enlarging those.)

The problem here is human, more than technical. Your boss has a particular display problem in mind. He/she thinks that the solution is an impossibly small initial image scaled to an insanely number of pixels. We know that's going to look awful, and you don't want your company or your boss to look bad because that won't help anybody.

My response, when I get something like this from a client, is to talk over the actual situation, find out what problem they are trying to solve and then come up with a solution that will work. Make no mistake, you can only do this with a live conversation. Memos, texts, emails are not likely to work. This also requires, on your part, complete sincerity, tact, creative thinking and persistence, in about that order of importance. Sometimes you have to work through some huffing and puffing from the other party, who doesn't want to look foolish or ignorant. (That's where tact and your own sincerity of purpose to find the best possible solution will help enormously. When the other party realizes they're not being made wrong, they don't have to insist on being right.)

Do what Marc suggests and run some tests with a portion of the image so you have something to "show and tell." But don't use this as a club to beat the boss into submission. You have to be on his side.

Keep asking questions until you completely understand the challenge your boss is up against, then come up with the right solution. If you make it seem like his/her idea, so much the better.

Answer 3

Off the bat..

1. Place image in Adobe Illustrator
2. Use the Live Trace feature to convert the photo to vectors
3. Scale all you want.

It's best to import the photo as large as possible to get as much detail as possible with Live Trace. And Live Trace tends to work better if photos are not exceptionally intricate. But it's a possible solution.

Sticking with rasters....

1. Upscale in steps.
2. 150% at a time until you reach the desired size.

You may benefit from applying a touch of unsharp mask at various stages.

Answer 4

An alternative to Illustrator's livetrace could be a halftone-like effect. The Size By Luminance script for Illustrator (look at the colour circle examples at the bottom) takes pixelated raster images and turns them into stylised, potentially full-colour halftone-like vector images that will then be scalable without looking accidentally pixelised.

If the image is neither wholly light nor dark you could maybe overlay the 'size by luminance' version over a heavily blurred version of the original, rather than a black or white background as suggested. A blurred raster, opacity around 75%, behind a 'Size By Luminance' layer, opacity around 75% opacity with blend mode set to 'Multiply', could be a good balance of looking nicely normal and photo-like from an intermediate distance, while not looking like a mistake up close.

You can adjust the relative balance of the blur layer against the dots layer based on how sharp you want the image to be (more dots layer = more sharpness and more contrast, more blur layer = softer focus and more accurate colours). If you wanted a pontilism-like effect you can probably get one by adjusting the dots and increasing their sizes.

This basic experiment ('On blur'), with minimal tweaking and adjusting, looks robustly non-mistake-like up close, and remains pretty accurate when scaled down to crudely simulate distance. The colours have changed a little, but I'm sure some tweaking could fix even that.

Worst case scenario: I'm sure with some work you can produce something using this technique that looks like a questionable but plausible design decision, rather than a mistake!

Answer 5

Viewing distance is incredibly important. It's incorrect to assume that all images for printing need to be 300DPI. If you find out the line screen of the final printing process, that will give you a hint to the maximum DPI needed for great reproduction.

Info about line screens of printing halftone patterns on Wikipedia

Why so big?

What's the use case for the 30.5 × 10.36m print? Is it a for a conference wall, where people will be walking right alongside it? If so, you'll need to do what you can to get the image looking great. Is it a billboard or poster than will be several meters away from the viewer? If so, pixelation may not be as much of an issue — a long time ago I created a billboard image with a (professional) 5MP camera. Pixels matter, but sometimes quality of the source is more important.

Testing

The best way to test all this is to create a sample portion of the poster at the right size, get it printed using a similar technique and see for yourself if it's acceptable or not. You can sell it into your boss as a potential cost saving, because large format prints are expensive, so mistakes with large format printing are a bad idea. You really only want to print the final poster once.

I've worked on lots of billboard advertising, but I don't think I've ever needed a 86400×29376 pixel file. That's pretty big!

Talk to your printer

Your printer should be able to point you in the right direction. I'd suspect you don't need to target such a high DPI though. In fact, it's typical for large format projects to be set up and 1/2, 1/4 or 1/10th scale in InDesign (or the app of your choice). So your printer might advise 150DPI at 1/4 size, or similar.

Techniques for scaling

If you do have to scale the image, I'd recommend using nearest neighbour and scaling to an exact multiple (200% or 300% etc). Or, try the other different scaling methods and see which you like the look of the most for the image you're using. Once you've done that you can add a little bit of noise (which will be done at the new resolution) and/or blur specific portions of the image, keeping complete control over how each area looks. This is what I did with the 5MP billboard image I mentioned earlier.

Don't trust any app that says it can magically add detail to an image. They can't. Try as hard as you can to ensure the source images are as high quality as possible. If the boss's sister's boyfriend took the photo on their iPhone, gently and calmly request that the photo be retaken with a better camera and/or by someone who knows what they're doing.

Answer 6

You can use Blow Up's trial > http://www.alienskin.com/blowup/blowup_example-3.aspx

Don't forget to buy it if you like it. It's ONLY $199 ;)

Answer 7

I don't know if its even still available, but I had great success with Genuine Fractals some years ago. I once made a banner that was designed to be viewed from two feet that was 4' tall by 10' wide, coming from a 1200 SPI slide scan -- a fraction of what you're trying to do, but GF was up to the task, with no visible pixellation.

GF works by describing the small scale image in mathematical fractal equations, then applying a scaling factor. This is similar in many ways to vector graphics, but it tends to describe photographic subjects better, whereas vectors tend to describe graphic arts (smooth curves, straight lines) better.