19

I want to enlarge an image to a bigger size. This will cause pixelation.

So I want to apply some treatment to it in Photoshop that will suppress some of this pixelation. Of course I cannot generate new pixel information, only smoothen it etc. so that the pixelation is not so much apparent in full views and prints. Example, one effect popular with photographers is to copy the image into a second layer, set its blending mode to overlay or soft light etc., then apply a blur to it. This somewhat smoothens the image. But it also changes its appearance into a glowy kind of look (which is the aim, usually).

I want a treatment that will fake the appearance of a higher-res smooth image without too many side-effects.

Any tricks of this kind?

17

If you've tried enlarging in Photoshop, the first thing is to experiment with the resampling algorithm (photoshop suggests bicubic smoother as the best for enlarging, but I have found it to be image dependent (if you have an image with a lot of edges vs a portrait or landscape).

Rather than smoothing, blurring I would suggest using a denoise program next, because they are smarter than any layer tricks, and can denoise while retaining sharpness.

Or you can use a product that is made specifically for this purpose, like Genuine Fractals. It promises up to 1000% enlargement without loss of quality.

Jeff Atwood has a blog post about this here: Better Image Resizing

  • 2
    I love this comment from the article: "if I really needed the Lena image that large, I'm better off hunting down old copies of Playboy and scanning them myself. Or at least that's what I tell my wife..." – fortran Jul 12 '13 at 14:56
  • 1
    Based on a few (eye) tests I've made, waifu2x (website, GitHub) is better than Genuine Fractals (or ON1). And waifu2x is also free. – Kenny LJ Jul 12 '17 at 23:56
13

I have found the free waifu2x very good for upsizing images. You can try an online demo. It uses "Deep Convolutional Neural Networks" to predict what the missing image data should be. It works better for line art, but is definitely acceptable for photos.

  • 2
    I'm normally skeptic on resampling programs, but this one make a decent result on some tests I made, inclusive on jpg noise removal. – Rafael Jul 10 '15 at 20:31
  • The only downside I see is that you need a very specific environment to install your instance and do the training of the neural network – abetancort Nov 14 '18 at 21:54
2

Based on this research paper, "Super-Resolution From a Single Image", from 2009 I found an implementation called QE SuperResolution which is an ok implementation.

The examples on the above page are just stunning (click the "SR" button to see the super-resolution version), and I doubt you could find anything better than that as of 2015.

2

The best techniques for enlargement of a pictorial scene involve vectorization and/or quantization in the frequency spectrum. One successful approach involves converting the image to a series of fractals. The fractal image is then resized. Some software, including plugins for photoshop and gIMP will convert your raster graphic to fractals, resize it, and then return the image data to raster form. It is worthy to note that software which upsamples in vector or frequency space will also apply some filtering before and/or after rasterization occurs.

I'm not inclined to hawk a particular product but if you google "fractal based image resizer" The results will give you several paid and open source options.

0

I use this website upscalepics to be very valuable. They use neural networks based on artificial intelligence. It basically predicts and puts some details to your photos with high accuracy making them look high resolution.It is optimized for photos and you can get pretty good quality pics up to 8x upscaling.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.