Is it possible to Colorize old photos with Artificial Intelligence or AI
With new developments in computer power so comes AI. This raises the question Is it possible to Colorize old photos with Artificial Intelligence or AI? Let us delve into what is currently available and what to expect in the futre.
Basic object colouring.
If you are an artist and want to draw but no colorize, there are some tools available we can use. A website called PaintsChainer uses AI to help color line art. Many artists draw cartoons and line based artworks. The tedious part for some is colorizing the art. I uploaded a line drawing art piece to their website to see how it works.
With PaintsChainer you can use the brush to help the AI paint in particular colors. You can choose between 3 styles and for an “artistic” interpretation it works quite well.
Using a public domain photo of a teddy bear converted to black and white i tried again. Not such a good result. Let us take a look at what is on offer colorize photos.
AI coloring for photos
Photographs are more complex than line art and require more complex solutions. demos.algorithmia.com offer a solution for colorizing black and white photos. I put it to the test. In order to get a fair comparison i used a photo i have already colored and the same image before it was colored. Below are some results.
You can see in this example the greenery has worked very well. Unfortunately the software uses just one shade of green and cannot distinguish between different types of foliage. It cannot identify the types of plant and therefore inserts “Green” as a colour. It may have recognized a brick pattern in the background and added “Brown” and has managed to pick the correct color for the ground. Most surprisingly this is almost the same color I used and the colour does not spill outside the confines of the walls.
On closer inspection the color spill is more evident.
If you used this for coloring a photo there would be so much clean up to do. Time correcting errors and spills would be better suited to using well known coloring methods.
What other AI colorizing is there?
Google Photos. Google have commented on and showed videos of their “Google Photos” app with a in built feature that will “colorize” old black and white photos. So far though i have not seen this feature included in any existing version of the app. The videos show it in action (Google – “colorize with Google photos” ) to find all there is to read. We can look forward to trying this out when it is released soon. Much talked about in May 2018 to be released in a few months, but not seen it yet!
Other offerings are out there but so far all colors spill out into each other with ill defined edges.
What is wrong with AI
The trouble with AI coloring is choice. AI does not make choices we humans can make. For example it appears to detect bricks and makes them red or brown. Depending on what geographical location the building is, the brick varies in colour and type of materials. What if the bricks are damaged or stained or have moss or lichen growing on them. This presents even more of a complex situation.
We have already seen how difficult it is to colour foliage. Even more tricky is to pick the correct lighting and to pick the scene apart to understand where the light is coming from, what time of day it might be by shadows and change the tones and color of light accordingly.
In the futre AI has a long way to get past identifying faces and shapes of bricks. Species of plants and therefore their correct colors. The face of person and their skin colour. Where one object stops and another starts.
Historically correct colorizing.
Even more difficult is historic colorizing. How can AI research the photo to find out what era, then what location? From here what colour object should be based upon know information and known facts. Ultimately choosing what colour to use. “Choosing” or is it programmed?
What will the future bring in AI colorization?
Is possible Colorize old photos with Artificial Intelligence? The answer is yes but there are many caveats. AI have much to learn and much to achieve.