Can active infrared recording produce original colors?

[SpaceBear]

Say you have a light source that produces light in the 700-1000 nm spectrum (Infrared), you use it in a dark room, and record with an infrared camera sensitive to the same spectrum.
The visible light spectrum is 400 nm (violet) - 700 nm (red). Therefore if you translate the recording into 700 nm = violet .. 1000 nm = red, then how realistic the colors you obtain will be?
I have seen only grey shades on active infrared pictures and I wonder how they would look in colors.
Has anyone tried that? Even if the resulting colors would be totally unrealistic, I think it's still something very interesting to contemplate.

The question came to me after reading http://en.wikipedia.org/wiki/Night_vision
and seeing the pictures
http://en.wikipedia.org/wiki/File:Extreme-CCTV-Active-Infrared-Night-Vision.jpg
http://en.wikipedia.org/wiki/File:Active-Infrared-Night-Vision.jpg

If the colors don't translate weel from 700 nm = violet to 1000 nm = red
maybe they translate better from 700 nm = red to 1000 nm = violet
or there can be many other forms of translating the colors. Maybe one such translation can produce the real colors?

 

[Drakkith]

What real colors? Beyond the visual spectrum there are no more colors so you cannot talk about "producing the real colors". You can assign colors to be displayed for various wavelength ranges, such as Red for 1000-900 nm, Green for 900-800 nm, and Blue for 800-700 nm, but this in no way would be "real color". It would be "false color" images. Remember that color is a perceptive feature that exists only in our minds. Wavelengths and frequencies are real, as we can measure them accurately, but we cannot measure color because it is not a real physical effect.

 

[MikeGomez]

The visible light spectrum is 400 nm (violet) - 700 nm (red). Therefore if you translate the recording into 700 nm = violet .. 1000 nm = red, then how realistic the colors you obtain will be?
I have seen only grey shades on active infrared pictures and I wonder how they would look in colors.

Even though the light source range of the active-infrared- night-vision (AINV) range is from 700 – 1000 nm, that doesn’t mean those frequencies were encoded in the photographic image. Looking at the images, it appears that the AINV helps enhance the intensity or brightness of the light only. It seems the color information lost, and the result is a purely black and white image.

Therefore if you tried to do what you suggest, the result will look more like an infrared heat signature image.

To get any realistic colors out of the image you would have to try to colorize the image as you would any other black and white image.

-------------------------------------------------

Having said that, there is a fascinating effect as shown in this video…



A photograph taken with a black and white camera through a red filter, is made to show vivid colors by viewing the photo through the red filter! Remember, the photo was taken with a black and white camera!

Edwin Land, who pioneering the Polaroid camera in the 1940’s, subsequently did valuable research in human color perception.

It’s a great thing to study, but my main point is that it seems plausible that there may be a way to recover some or all of the color information of the AINV image. Just an idea though. I could be wrong.

 

[SpaceBear]

What real colors? Beyond the visual spectrum there are no more colors so you cannot talk about "producing the real colors". You can assign colors to be displayed for various wavelength ranges, such as Red for 1000-900 nm, Green for 900-800 nm, and Blue for 800-700 nm, but this in no way would be "real color".

True. And assigning in the reverse order you would get a different false color picture. And playing with the spectrum (for example: starts with red, ends with yellow) will give another false color.
The question is: what if any of these pictures taken in false color are similar or very similar with the ones in true color?

It seems the color information lost, and the result is a purely black and white image.

True. In order to experiment this, the camera must not lose the color information, and has to work the thermal infrared cameras, which show colors instead of B/W

Therefore if you tried to do what you suggest, the result will look more like an infrared heat signature image.

That's possible but I also think it's more probable to look similar with Ultraviolet photography: http://en.wikipedia.org/wiki/File:UV_Portrait.jpg

So the easiest way to experiment with that would be if someone takes the picture in visible light so we know the real colors, then it takes the picture in near-infrared (without losing color information), and then release it to the public so people can try to assign colors to different wavelengths. Maybe if you create a spectrum that starts at 700 nm with green and end at 1000 nm with yellow would be the solution. Therefore the people can play with new color spectrums, defined by start, end, and intermediary colors.

It’s a great thing to study, but my main point is that it seems plausible that there may be a way to recover some or all of the color information of the AINV image. Just an idea though. I could be wrong.

I totally agree. Even if the results wouldn't be much, it would be an enriching experience!

And the same thing applies for Ultraviolet photography also.
Someone should create a Kickstarter platform for such distributed research i guess..

 

[mikeph]

This would only be possible with a complete calibration, based on knowledge of your subject's emission spectrum. Since this spectrum changes based on temperature, material, etc. the knowledge is very much limited, and your calibration becomes just a guess.

 

[Drakkith]

I see no reason to assign colors to part of the EM spectrum that simply doesn't have colors. It is incredibly easy to make false color images. It doesn't require highly specialized equipment either. Many digital cameras can be modified to see infrared by using simple filters.

http://en.wikipedia.org/wiki/Infrared_photography

 

[mikeph]

As I understand it he's not trying to assign colours to IR specra, he's trying to use the IR image to predict what the visible image would look like.

If everything in the room is a black body this should be possible, you just fit a blackbody spectrum to each pixel and extrapolate into the visible light to assign "real" colour. My point was that things are rarely blackbody, so it's hard to achieve in practice.

 

[Drakkith]

As I understand it he's not trying to assign colours to IR specra, he's trying to use the IR image to predict what the visible image would look like.

If everything in the room is a black body this should be possible, you just fit a blackbody spectrum to each pixel and extrapolate into the visible light to assign "real" colour. My point was that things are rarely blackbody, so it's hard to achieve in practice.

Even if everything were a blackbody I don't think you could do it, as at room temperature objects do not emit significant amounts of NIR (Near Infrared) light. That's why most night vision cameras have their own IR light sources they use.

 

[SpaceBear]

I think I have to formulate it again:
Imagine you take a photo in Near Infrared (NIR) - let's say in 700-1000 nm. If it's in the day, then the scene is lighted in NIR, because the sun emits also infrared light - so you don't have to use your own light source. If it's in the night, you have to use an artificial light source to light the scene, in the full NIR spectrum used (700-1000 nm). Using light bulb emitting light from 400 to 500 nm will make your picture taken in visible light to look bluish, and the same applies to a partial NIR spectrum light - the picture will look coloured by the partial spectrum (for example 700-800 nm). So the light has to be in the full spectrum you work in.

When you take a picture in visible light, you can tell the wavelength of the light emitted by the objects in the pictures: a red tomato reflects light on about 700 nm, an orange reflects on 600 nm, a green leaf reflects on 500 nm, and so on.
Of course the wavelength is not encoded in every pixel, but it can be deduced, from the color of each pixel (in visible light photos).

So the question is how to take a photo in NIR in a way that you can tell the wavelength of each individual pixel, like you can do for a photo taken in visible light.
Because from the NIR photos shown in the Wikipedia page it seems the pixels record only the intensity of the light. (how much NIR light the objects reflect)

A leaf is green to our eyes because it reflects the green light (500 nm). But maybe it also reflects other wavelengths too, in the NIR spectrum, and in other spectrum.

 

[SpaceBear]

As I understand it he's not trying to assign colours to IR specra, he's trying to use the IR image to predict what the visible image would look like.

By the contrary, I am trying to assing colours to NIR spectra (700-1000 nm). Did anyone try that?

Reconstructing the colors from a black and white image taken in IR is also an interesting thing but I'm not sure how much it would be possible, as it involves too much guessing.

 

[Drakkith]

Look up what a bayer filter or bayer array is. You could do the same thing to IR light that you can for visible. Note that all digital sensors only record the intensity of the light falling on them. They are completely unable to tell what wavelength the light is.

 

[SpaceBear]

Look up what a bayer filter or bayer array is. You could do the same thing to IR light that you can for visible. Note that all digital sensors only record the intensity of the light falling on them. They are completely unable to tell what wavelength the light is.

Yes you can. If a pixel is red, then the wavelength is closer to 700 nm than to 400 nm. If if's purple, then it's closer to 400 nm and so on. So you can tell the wavelength. There are also devices that can measure the wavelength of an EM radiation, including light. The wavelength is not encoded in the pixel information directly but it can be deduced very accurately from the RGB information.

The camera sensors have 4 individual sensors for each pixel: 1xRed, 2xGreen and 1xBlue.
Therefore if you shoot in NIR (700 to 1000 nm) you must have a different type of CCD, that records the intensity of other wavelengths instead of RGB. Say you have a CCD that records the intensity of 1000 nm, 800 nm and 850 nm (the corresponding of RGB into 700-1000 nm). Based on those three components, you can tell the wavelength of each pixel, even if they don't mean any color for our eye.
And when you have a photo taken into 700-1000 nm NIR with 3 bytes per pixel, you can play assigning colors to the 700-1000 nm spectrum.

 

[Drakkith]

Yes you can. If a pixel is red, then the wavelength is closer to 700 nm than to 400 nm. If if's purple, then it's closer to 400 nm and so on. So you can tell the wavelength. There are also devices that can measure the wavelength of an EM radiation, including light. The wavelength is not encoded in the pixel information directly but it can be deduced very accurately from the RGB information.

The COLOR is, but not the WAVELENGTH. You can say that it is *probably* within a certain range of wavelengths, but that's it. I don't consider that as a reasonable definition of "accurate".

The camera sensors have 4 individual sensors for each pixel: 1xRed, 2xGreen and 1xBlue.
Therefore if you shoot in NIR (700 to 1000 nm) you must have a different type of CCD, that records the intensity of other wavelengths instead of RGB. Say you have a CCD that records the intensity of 1000 nm, 800 nm and 850 nm (the corresponding of RGB into 700-1000 nm). Based on those three components, you can tell the wavelength of each pixel, even if they don't mean any color for our eye.

Again, replace wavelength with color and you are correct. This would make a false color image of the NIR band.

 

[reasonableman]

I believe Drakkith is hinting at it but I'll say it explicitly, to make a 'infrared colour image' get three filters e.g. 700 - 800, 800 - 900 and 900 - 1,000 nm. Then take an image with each filter in front of the detector - a total of three images. Then use each image as an R, G or B element of an image i.e. combine them to form a single image.

This is how color photos can be taken with a B&W camera. Of course in the infrared your filter choices are completely arbitrary.