Have a thought for image restorers

Hello all, I hope everyone is keeping well.

My first, and yet only Topaz product is DeNoize, and I love it. I tried Gigapixel, but it never helped with many of the poorer quality images I usually end up working on. Yes, if I am working on a Hi-res photo from a 5 year old camera and want to bring it up to date, yay, sorted. However, if someone has an image taken from a 30 year old slide, and only got given an image at 200 dpi and now they have lost the original, Gigapixel would more likely make a mushy cartoon image out of the original that doesn’t improve anything.

I tried the video tool, and it did nothing for a 1930 newsreel film.

I would love a tool that could help remove text haloing in over-sharpened images, I would love help with Jpeg blocking. To use an analogy, these products are great in getting a scuff out of your car paint job, not so good if the car has a prang, dent or chip.

Hello Noblelox,

Image restoration is a very complex process. This is not a 1-2-3 process. This process is so difficult that, for example, specialized tools for the restoration of old films are really very expensive and ahead of light years what is offered by, for example, Photoshop.

I mean automatic and semi-automatic tools.

Because manually every computer graphic designer can get rid of distortions from photos using any raster graphics editors and his own ingenuity. Only this is not time effective like these specialized tools, you need a lot of time to do it.

However, a semi-automatic image restoration, not only of movies, but also photos at home for amateurs, at a very high level is a real challenge if you are not a large company and you do not have as much money for these tools as 10,000 - 100,000 USD.

What can you do at home if you are not a millionaire?

For example, you can create a blur effect with a radius of 0.5 px. This can remove “halo” from some images very sharp. You can experiment 0.3 px - 1.2 px etc. But this is not a remedy for all artifacts of this type.

There are FFT filters for the image (based on a two-dimensional Fourier transform) that break the image into a 2D spectrum. Here you can do more and eliminate the raster or Moire phenomenon effectively. Also various distortions.

You can also use a neural network in the Pix2Pix (AI) model and train on tens of thousands of image pairs. One image is normal, the other is extremely sharp. After some time, the neural network will be able to remove halo from new images.

For this you need a graphics card containing effective GPUs. Learning lasts from several days to a week, depending on the number of photos. I did it myself, because I am also an AI software developer. But only for own purposes :slight_smile:

However, you may have trouble collecting enough high-quality photos from sites like Pixabay.com, because most of them are edited and sharpened, so it can be distorted.

Another problem is mastering and learning all this technology. I am a programmer so I have no problem, but an ordinary photographer or graphic designer will have difficulty installing all program modules, e.g. from NVIDIA, configuring the computer environment. This may discourage most people from experimenting. Not to mention that you need to learn to program. (The latter is not necessary, because there are ready implementations of the algorithms). Anyway, to start doing something like that, you also have to learn many things, so you have to invest a few months of learning, at least in the basics.

Topaz Labs AI filters work on a similar principle, such as: Denoise AI, Sharpen AI, Jpg2Raw or Gigapixel. Only they offer ready-to-use tools, so you don’t need to know anything about AI :slight_smile:

Unfortunately, there is not yet a tool for comprehensive image restoration for less money, based on hybrid AI algorithms combined with FFT, object recognition, Bayer nonlinear filters or wavelet transform. There is more of it.

Indeed, Topaz could make another AI-based product, a product to remove the halo effect from HDR photos and from too sharp photos, the printing grid, dirt and other things. And even for subjectively improving photos according to e.g. painting style (something other than classic Neural Style Transfer).

Overall, there are hundreds of ideas for new products. And some of them can be done. Unfortunately, there is another problem: who will do these tools and then buy?

Topaz operates under the conditions of capitalism, so it must focus on what is profitable. The company is now doing things that everyone is doing, so: sharpening, denoise etc. It is useful for professionals, so there is business.

In contrast, unusual and fancy, but at the same time cheap tools for creative people at the level of 100 - 200 USD, using new technologies may never arise, because there will be too few dreamers who would buy these unusual tools. Tools with unusual filters, with an unusual interface …

Imagine that instead of turning on Photoshop, you turn on the Topaz Fantasy Creator product and simply type in the box (or even speak into the microphone):

- TPS, remove this cloudy sky from the picture and replace it with the sun, please. Then paint me a girl with a dog. Use a straight line. Then remove every other car from the picture and replace it with a tree. Add more flowers in the background.

And the system will create a drawing-photographic hybrid in a few minutes. No tedious editing. This may be the future of artists. Because artists will still be needed, because someone must have imagination to give joy to others.

Lech Balcerzak