I was recently asked to try to “restore” the colour in a much faded image, which I was happy to investigate. But I am not an advocate of the retouching software, whether it is Photoshop, Capture One, FastStone, Affinity Photo, GIMP or one of the other 155 different variants. Being an advocate of form over colour (i.e. I prefer Black and White), I have never bothered to learn the intricacies of such techniques. Part of it is because I prefer to think of myself as somebody who creates in the viewfinder and not compensating for a lack of content through the retouching (ouch!).
But having said that, I did want to investigate something I had seen at some point though not sure where, namely the use of AI to re-insert “original” colour. Life is too short for me to spend time looking at how it is done, but I guess there are a couple of possibilities. A) to find similar images (using something akin to when you look for a similar image in Google Images), and use that as a guide to the palette that should be used, and b) to look at the existing palette in the image and make best guesses (probabilities based on experience) about how to “re-saturate” the colours in a manner that makes sense. (My guessing is limited. If you know, let me know.)
So I looked for a photo that was suited to the themes of this blog, and was pleasantly surprised at the outcomes. The original is a 35mm slide from around 45 years ago. Not surprisingly, iti is a little faded and lacks some of the original colour. Given that I do not know what the “true” colours should be, the evaluation is obviously highly subjective. But I would say that this is clearly a good start in a field that I am sure will advance over time.
That said, I have decided to only show one of more than a dozen alternative suggestions. The site is paletle.fm, and it offers 20 alternative palettes. The reason I chose this one was that it was free, though the output size is limited to 408 x 612 pixels from an original of 1680 x 2520 pixels. The only other rendition worthy of consideration for this image was “Vivid natural”, which gave a very green grass.
Within the image in Palette, there was a hint of how it achieves the results as I found the the system appears to analysed the image and provide keywords, specifically “A stock photo of flesh, ape, limb, hen, body, nudist, and mildew. Black and white.” The Base Palette resultant image was classified as “A stock photo of cheek, flesh, nudist, and female. Small contrasting details in color,. In color film.” So a little more human!
There was not time to investigate everything. But two small hints of things I may investigate in future. Firstly, when I re-uploaded that image, the colours were slightly different, and I would say slightly better. Random? Related to cached image? Who knows. But where it gets even more curious is that when I upload that same image as a black and white, the colours were remarkably similar. (Note that “black and white” also appeared to be a keyword of the original.) A good outcome, but a slightly spooky result!.