Adding colour to vintage Wolves shirts – colourisation

A look at how we can see those old Black & White photos of Wolves in a whole new light… with a controversial punchline

In this page:

Adding colour to photos

Adding colour to video content

Adding colour to photos

I made someone cry the other day – not intentionally and in a happy way I hasten to add, but because of something I did to their old photos. No I didn’t lose them – you see it’s not just Wolves that I’m mad about, I’m a bit of a tech geek as well and one topic that has really interested me of late is Artificial Intelligence-based colourisation (or colorization if you are of American persuasion) of vintage B&W photos. Long story short, because this is a Wolves and not a tech blog, old photos may be black and white, sepia or monochrome but if:

1. They are of decent quality
2. Are lit well (but not too well as shown in a moment)
3. Are scanned at a good resolution

… they do often still contain enough of an indication of what the original colour was for a computer to analyse and re-build it. A germ of the original colour if you like. Please note that this is not the same as the painstaking restoration and Photoshop restoration that some talented people do which is more akin to art I think – this is pure machine logic…Want an example?

Interesting isn’t it? It’s not an exact science and it is not perfect (note some of the colours on the lady in the middle’s coat) but I’ve been told by some of the subjects of the many photos I ran, that the clothes colours are accurate to their recollection.

Of course next I needed to see how it works for old black and white Wolves photos. Cue – me getting my old books out and getting scanning! Let’s look at a wonderful old book called ‘The Wolves’: The First 80 Years by Percy M Young. There is a photo on one page showing Wolves squads in 1905 and 1920:

An now, for the first time since the photos were taken, an enhanced view with the old gold peeking out:

Wolves in colour… for the first time

Fascinating but not perfect. Works well on the 1905 kit which featured a deeper old gold but not so well on the 1920 more saturised image, although you can see the faint bronze tone trying to rise to the surface. Long story short, I found that brighter (mainly later) Wolves shirts which are lighter in tone don’t work as well as the older old gold – the AI has difficulty distinguishing brighter orange shades from blue due to where it sits on the spectrum.

Does work very nicely on a certain Ron Flowers playing on the Wembley pitch (note it got the England badge wrong, should be black):

Even with low-resolution images taken from other old books you can see the original gold colouring under the surface. You could probably enhance the photos afterwards and make more obvious but felt like that would be ‘cheating’:

Or our captain training perhaps?

I think that the additional colour, even with the imperfections, adds a whole extra dimension to the photos and makes them feel more ‘real’.

Where I spent most of the time however was on scanning, cleaning up and also enhancing one particular black and white photo, again from Mr Young’s book which purports to be the (at the time) earliest-known photo of the Wolves team from “almost 70 years ago”. The book was written in 1959 so that would make the photo from 1890-1891 or 1891-1892.

Fair enough I thought. But let’s have a closer look. Firstly this is the cleaned-up, enhanced and smoothed version:

…and here it is having been saturised and run through colourisation process:

OK two things to note, and I may get shot down in flames for this.

The blue home shirt of 1891 myth

I know that some people do have the opinion that Wolves did very briefly have blue shirts at that time and that they almost immediately swapped them to half and half diagonal black/old gold (as backed up by reports of the time). That at least is evidenced up by this colorised shot which does suggest the shirts are blue of varying shades. However I do think some people may say this based purely on the evidence of this one early photo.

However – take a good look again at some of the players. Here is an enlarged view of some of them:

The book was published 60 odd years ago and since then earlier photos of Wolves teams have emerged including:

• One from circa 1883 on the Football Network site
• Another recently shared by Wolves themselves showing the 1889 cup final team (two years prior).

Look at both of those photos: none of the people are the same. Further, the other photos show the handlebar moustaches and stiff upper lips of Victorian gentleman footballers in their prime. You don’t see that on this 1890/91 photo because these are kids. Smaller in stature than the older gentlemen stood besides them without a wisp of facial hair between them.

I can’t prove this either way but I do not think that this is a photo of the Wolverhampton Wanderers team as stated in this book. I think it may rather be a photo of the continuing St Luke’s school team (a theory which fits with their old school colours) or another local, associated team.

Otherwise… explain

• How the average age of the Wolves team tumbled in the space of a few years
• Why the players all changed at a period where transfers to other teams was not commonplace
• …most pertinently how these lads were not kicked up in the air against let’s say, for example, the Villa team of the same period:

I think they didn’t because they never would have played them… because that isn’t Wolves in that photo. Happy to be proved wrong…

Adding to colour to video content

A gent (I assume he is a gent) called Olivergoldblack on the forum Molineux Mix has been doing something very similar but with video images as opposed to photographs. His results so far may be viewed here. We have asked that he keeps us updated and we will update this page as and when this progresses.

In his own words:

I’ve been playing with computer vision to colorify black and white images. The algorithm that converts the colours ‘hallucinates’, what colour the black and white pixel colours should be. It guesses this off what it learns off the training data. The training data is colour images converted to black and white, so the algorithm learns the conversion. So when you feed in native black and white images, it does its best guess.
The more specific the training data the better the colour guess would be. I.e. you’ll always find grass, sky, roads are all pretty good in this process as there’s loads of examples, but colours of clothes is much more difficult, as there is so much vary.