I’m probably a brainless jacket potato asking this question but it has always intrigued me. Recently there was some footage posted with an old lady making thread or something and I’ve seen the footage get upscaled and colourised etc and I’m wondering how, if the footage is black and white, it can be transformed.
Side note:
Thank you all so much for your answers and comments! Also a massive thank you for my first ever award! Very much appreciated honestly 🙂 This really lends perspective on how much work people put in to colourising old videos!
Side note 2.0:
Thank you for the second award! Never thought that my curiosity would net me all this information or awards! You guys are all champs 💪🏻
In: Other
Can they use specific shades of gray to automatically translate to a certain color? That’s how I always assumed they did it. But I guess not
Three ways:
1. By manually coloring it in using Photoshop or even actual paint.
2. By teaching an AI to do it.
3. By extracting some color information from the “noise” in the signal. Sometimes this is possible, depending how it was recorded.
Oh man. I was intrested in coloring old photos so I watched a tutorial on youtube. “It cant be this easy!” I said to myself, all you have to do is guess what colors the different parts of the photo should be and then have the overlay setting on. The shadow of the picture does the rest!
Was writing this as a reply to someone else’s comment but I figured I might as well reply to OP.
True ELI5: Huge numbers of people, usually outsourced to poorer countries and supervised by a small team in the business’s home country will trace areas of the footage frame by frame and apply color with vfx editing software – think Photoshop, but for video. The colors are not necessarily correct, just what the supervisors think it would have been.
More detailed original comment: Up to hundreds sometimes. Lots of it in India. The same thing with stereo 3d conversion. I used to work at a company that turned movies into 3d by having large teams rotoscope hundreds or even thousands of individual objects/limbs/environment elements which got processed by other artists to do the actual conversion. The tools we used had some techniques for tracking some objects and bodies, but it was a very manual process. When I first came to the company I helped out with roto for the first month. On one hand, it was grueling. On the other hand it could be relaxing. I’ve sort of gone on a tangent talking about roto in conversion since that’s my experience, but it’s roughly the same idea for colorization for separating elements.
Lastly, as I type this I should note that as a compositor, there are some tricks that could be used. Consider a plaid texture shirt. You wouldn’t want to roto every single line to colorize it. Instead, you could roto just the shirt in its entirety and use its luma information to map color to the patterns. There a probably a lot of other techniques like this, but this one is the first that comes to mind.
1. Luminance from Black and white images can be calculated
2. Use those assumptions and then let artists colour stuff with some knowledge of how stuff colour should look. Ex. Colouring a Taxi yellow or colouring a tree green. Just a series of reasonable assumptions by the artist. If a lot of man hour is poured into the project then the result will be pretty evident since you’ll see less “out of the ordinary” coloring. Unless sometimes they look odd if colour assumptions are taken a lil too liberally. For example, In my family we have a coloured portrait of my Great Grandpa and Grandma, was monochrome and it looks all right but if you focus you can tell skin colour is off. And people who say them can tell very easily what’s wrong with the coloring.
Machine learning or spotting the artifacts the camera produced for each color which sometimes can be used to find the color.
Just like coloring books … and make sure to stay in the lines 😉
No smell of crayon wax, but you do get the hum of a computer fan while doing it.
I feel like if you take a lot of photos of different colors and materials and turn them black in white it would help to make it easier when adding color to old photos.
To explain like you’re 5. It’s like a coloring book. You have your Frozen coloring book. You know Elsa’s hair is blonde, her dress is blue, trees are green, etc.
It’s the same concept except the experts who colorize photographs tend to do a little more research to try and get the colors and shades correct.
This one for example? https://www.youtube.com/watch?v=EQs5VxNPhzk
it’s done by machine learning. The machine gets fed color movies and the same movies with the colors stripped and has to learn how to make one from the other.
It actually learns to detect e.g. trees and that they are green. I love that in my example it can tell a statue apart from a human 🙂
Latest Answers