View Full Version : Colour recognition
17th November 2008, 08:31
Anyone know if there is available a program or technology that can recognise accurate colours from a b/w photograph.
If you fed in a palette of say 100 colours and hues of colours and turned them into a greyscale chart could it then be possible to recognise the original colours on a b/w photograph.
Some SN members will know I am building a data base of old trawlers company funnel colours and such a program would be an immense help.
A pipe dream or what !!!!
17th November 2008, 13:23
Since there's no colour information in a black and white photograph I doubt if you'll find anything.
The only things I know are ones where you assign a colour and the computer works out where to put it.
17th November 2008, 15:18
Oh dear, thatís not what I wanted to hear.
I had hoped that the computer was such a devilishly clever thing and that someone somewhere had worked out how to do this.
I was thinking about those million light years away galaxies that we see in such wonderful colours, do they just make the colours up or is that what they see and what they look like?
17th November 2008, 15:33
They do colour old movies with computer technique in Hollywood. I remember there was a lot of protest against it when it was introduced: tampering with the art of dead artists etc. From Wikipedia: In March 2008, it was announced that new colorization technology, which involves detecting colour artefacts ("dot crawl") in high-resolution scans of the black-and-white films, will be used to restore Doctor Who episodes, as well as shows like Steptoe and Son where some episodes only exist in black and white. A team at the Hebrew University of Jerusalem's Benin School of Computer Science and Engineering describe their method as an interactive process which does not require precise, manual, region detection, nor accurate tracking and is based on the simple premise that nearby pixels in space and time that have similar gray levels should also have similar colours. At the University of Minnesota, a colour propagation method was developed that uses geodesic distance.
But to me deciding what colour a certain percentage of black should represent sounds rather impossible, it is not as if red and blue could not reach the same level of darkness.
Intriguing question though, does red produce more reflex than blue at the same level of darkness? Does yellow shade less or more abruptly on a bent surface than does green? I am not ruling it out that a computer program taking into account every part of a perfectly sharp b/w picture could not, at least theoretically, make sensible suggestions… The suggested method I will, as does Marconi Sahib, rule out though. Regards, Stein.
17th November 2008, 16:52
What we are coming back to is the plain fact that th ehuman brain is more adaptive and more appreciative, especially in the area of interpolation of data to create the whole than any computer; the computer and any algorithm is based entirely upon a single or group assessment of human brains, and any action taken by a computer will always give that interpretation alone; I detest black & white movies being 'colorized' and I always turn the picture to monochrome to watch them if they have been mauled by these technicians with no appreciation of art and craft. As to dicovering colours on B&W photographs, I think this is down to having many pictures of coloured and B&W of the same subject and to make an educated 'guess' at any that are only B&W images. It depends on so many factors, that I feel it may be risky to rely on someone else's algorithm to provide the answers that you need.
For example, it depends on the time of day, the cloud cover, etc .. look at colours in St Ives, and they will be different from the same colours in Grimsby.
All the best, Raymond
17th November 2008, 17:04
You could try opening the photo in a program such as Paint Shop Pro and then inserting examples of funnel colour and houseflag as separate inserts in colour. Hopefully the example I show will indicate what I mean. The colours I have used are arbitrary and not real to the ship shown.
17th November 2008, 18:51
Something else that you could try is to find colour pictures of various funnel markings and to reduce them to 'grey scale' and see what the general tones look like in comparison to the ones on which you are stuck. Make sure that the daylight is much the same. A clear sunrise is vastly different from a midday cloud bank.
17th November 2008, 21:32
The colours in astonomical objects is not made up, but is actually there. It is however often enhanced to give a better idea of the structure. Many stars have colour, even to the naked eye (e.g. top left star of Orion, Betelgeuse is a red giant, has a definate red tinge).
The human eye is very poor on colour definition at low light levels, which is why when you see astronomical objects in photographs you may be a bit astounded by the colouring. But the light gathering power of modern telescopes, plus a little enhancement, plus filtering if you are looking for specific wavelengths (in order to identify specific atoms or molecules) brings out the marvellous colours.
17th November 2008, 22:48
Your question was is there a program that will add colour to monochrome photos. I presume you also want it to get the colours correct.
The simple answer is no.
A more complex answer is that such a computer program would need rules to tell it what colours to assign. The information to create those rules and get the colour right does not exist in the information in a monochrome photo. The best such a program could do is to work on generalised rules - e.g. sea is green, sky is blue etc.
18th November 2008, 10:07
Can I thank you all for your very erudite answers to my question.
I did comment in my original question as to weather it was a pipe dream to expect a positive answer and it would appear that it was.
It would have made life a lot easier for my project and who really knows what may come along in the distant future, but will I be here to see it and use it ?
18th November 2008, 12:19
Hello Bill, just to start you off ... here are two figures, one is original colour the other grey scale - you can see already the problems in assessing comparative colours, in light or dark shades, etc ..
18th November 2008, 12:33
It certainly looks impossible to our eyes. I had tried a similar exercise myself with the same effects but had hoped an "eyeless" computer may be able to differentiate between them but you will see from the other above answers certainly not at the moment. All good fun though and it keeps the aging grey cells busy as Hercule Poirot would say.
18th November 2008, 18:11
I certainly would not say impossible, our brains are more than pre-programmed and much more than learning machines .. we have an inbuilt assessor and adaptor, which will allow our minds to fill in gaps in wording, errors in spelling, put in missing letters or damaged letters ... With a lot of inspection, a series of prime attempts followed by re-assessment, I am certain you could get very close, if not accurate, answers. With all the range of shades and comparative darknesses, I have grave doubts about an algorithm to solve what we can do simply by using the little grey cells.
All the best, Raymond
vBulletin® v3.6.8, Copyright ©2000-2013, Jelsoft Enterprises Ltd.