Infrared Images with Canon 20D Conversion

[David]

Hi Richard - Great shots - I've not had any focusing problems with the EF and EFS lenses, and I think that the firm that did the conversion for me take that into account somehow. As I mentioned in an addition above, I've got myself some Super Takumar lenses that seem not to show any hot-spotting at small apertures. The problem there, however, is that of manual focusing. In principal, then focusing for IR become a problem, but so far I have not had an issue with it in test shots. Further, those lenses have the magic IR mark on the aperture ring. But I'll have to check it all out when time permits.

Many thanks for your interest. I'm following on with another post on this thread about spectral response. I hope that will provide further information.

David

[David]

I came across this graph the other day:



This is what's called a spectral response curve, and it shows how the red, green and blue sensors in a Canon 40D respond to different wavelengths of visible and near infra-red (NIR) light.

The source is: http://www.maxmax.com/spectral_response.htm

The company MaxMax (based in New Jersey, US) carries out camera conversions and has some interesting information about IR and visible spectra as well as several spectral response curves for different digital cameras, particularly Nikon.

My camera conversion is of a Canon 20D, but I expect the spectral response curve to be very similar as I doubt Canon changes its dyes and pigments for it filters every time it brings out a new model.

We can use this curve to explain a great deal about what goes on in NIR photography. First, note that for cameras fitted with an R72 filter the relevant part for NIR light is from 720 nm to 1000 nm. That is the radiation to which the camera responds, not to blue, green, yellow or orange light.

Second, note that the red sensor response dominates both blue and green sensors responses in the NIR region. This is why, if using automatic white balance, the images appear with a red cast. Some books and other Web sites suggest that it is leakage of red light itself that causes the cast, but not from this response curve. It's the red sensor's response to NIR light that matters.

Third, blue and green curves show that both these sensors have a response to NIR light, green is weak, but blue builds up around 770 - 780 nm. Prior to that there is only a very weak response by the blue channel around 690 nm. Now this turns out to be important because we can say that there is effectively no blue response before 770 - 780 nm. If we look at the following image we can see why:



This is a white balanced shot of an estate agent's sign. Note that very specific areas are pale blue, yet there is no blue light reaching the sensors. The original image had a red cast, but that has been removed by the white balance. The blue, however, has not. We can explain this with the aid of the spectral response curve because we know that the blue channel has a reasonable good response at and above 770 - 780 nm. Yet, this does not in itself explain why these very specific areas are coloured blue. In order to explain that we have to note that if the blue channel is responding, then the NIR light must be at or above 770 -780 nm and there cannot be any from shorter wavelengths. Why not? Because the sign in those specfic areas is absorbing that light. Now usually books etc. tell us that NIR light is reflected, but not absorbed. This image and the spectral response curve show that is not always the case.

The last point I would mention is that the response curve illustrates why NIR photographs can be manipulated via white balance, channel mixer or other means to provide such a rich diversity of outcome. It's the channel differentiation that allows these changes to be made.

(Strictly, just in case there is any confusion, when I talk about NIR photographs I mean the visible images that we can see after the camera sensors and the camera's firmware have converted NIR light to a visible output - a bit like an HDR image actually being a LDR image tone-mapped from an HDR file.)


Cheers

David

[David]

Hi All - Another few notes. Look at this image. How banal can you get! Boring or what! It's the backside of a pillar box with a standard grey card stuck on top. (Pillar boxes are red in the UK, BTW). If there were a competition for mundane uninteresting shots this would be a front runner - red cast, poor contrast, little dynamic range.




(No doubt, if GCHQ is monitoring this web site I'll be getting a visit from Special Branch for defacing Government property, taking photographs of a high security installation, and generally being a terrorist threat. I'm waiting for the knock on the door.)

And yet, this shot has one property that is difficult to attain in NIR photography – there is no clipping at either the white or black end of the histogram for any of the three channels, red, green, and blue. (There may actually be tiny amounts but of no significance.)

The next image shows a screen grab of the histogram of the above photograph.




From the previous post, with the spectral response curve, the dominance of the red channel in NIR photography was graphically shown. If you take an image with the Canon 20D, the camera shows the image on screen along with its histogram. However, that image is in JPG format and the histogram is a Value histogram with no colour information. If you expose for a reasonable histogram you will be clipping, possibly severely, the red channel. If you “expose-to-the-right” as often recommended for visible photography then the red channel will be hopelessly over-exposed. That would not normally be considered a good thing.

The photograph was obtained on a dull day, hence little dynamic range, and comprises large areas of grey, some greenery, a red pillar box, and various printed matter and text. The exposure details are Av 5.6, Tv 1/350, 18-55 mm EFS lens at 55 mm. I had taken 8 shots at different exposures to find this particular one. The camera metered shot was Tv at 1/180 and was over-exposed in the red channel.

Why should this be of any importance – dull photograph, dull post? (pun – geddit?) Yet, this image allows me to test some aspects of NIR photography. The first is to compare what information is contained in the different channels. If you have one channel dominant then comparing that to the other two is difficult, because they are then under-exposed. But here all three channels have reasonable exposures. In addition, in this image there are no colour artefacts as illustrated in the previous post. Individual channel images were obtained by conversion of the RAW file to a JPG, then decomposing (using the Gimp) to red, green and blue channels. The image channel images were then sharpened and a Levels correction made so all were of similar value. The next image shows the result for the red channel.





The important result is that all three channels contain, for this image, almost completely identical information. There are some tiny changes in relative intensities, but nothing that would indicate the green and blue sensors are receiving different input from the red sensor. (The red channel despite my efforts does show some small areas of clipping, but such is not present in the other two channels.) Why should this be important? Because it means that if you do blow the red channel then you are not going to lose information from the image. Indeed, you could consider that a blown red channel actually “subsidises” the other two channels, when a white balance correction is applied, by pulling up their values into a reasonable exposure range. Now this is not the case in visible light photography, because the different channels often (usually) contain very different information and blowing a channel can affect the whole image.

Thus, in summary, provided the red channel is not grossly blown, such as would happen if you “exposed-to-the-right”, a NIR image ought to contain all available information. Actually, this ought not to be surprising if you consider the spectral response curve. With visible light it is differentiated with respect to wavelength of light. However, in the NIR region, provided there is no absorption, the curve is differentiated by response not wavelength.

The second aspect of NIR photography that can be checked with this image is that of focus. As far as I can tell, having made sure that the camera did focus, in this case on the grey card, there is no obvious problem either in the overall image, nor in any of the three channels once typical exposure sharpening is applied.

Now the third thing we can do with this image is....... Ah, there's a knock at the door. I wonder who that could be? Back soon?

David

[David]

In previous posts I've shown how the camera response curve can be used to explain some colour artefacts in NIR photographs. This post explores some further issues of colour and NIR images. The principal subjects in each of the following shots have something in common: they are all wearing at least one substantial item of "black" clothing. As it turns out "black" is not all it seems to be.

All shots are taken with the converted Canon 20D fitted with the 720 nm filter. An automatic white balance had been set so that a red cast was seen in all the unprocessed shots. Exposure was kept as low as possible so that the red channel was not clipped, although as the previous post explained this may not be a problem. Post processing involved using a previously established grey card setting to give a neutral white balance. Images were then cropped to avoid extraneous detail.

This first image shows a young woman wearing a pair of "black" jeans. The image also shows a lamp-post as partially black and the woman's bag and watch strap are black. The blue cast of the jeans is in accord with a previous post in which I showed how absorption of very near IR light allows the blue sensor to come into play in NIR images. Notice also the woman's hair. It too has a blue cast - possibly hair dye is the "root" cause



So far so good, but the second image takes us further.



In this picture the girl's leggings are allegedly "black", boots and bag are but not the leggings in the NIR image. What's more, the man in the shot was wearing a black suit - apparently not matching jacket and trousers. Thus, we see that NIR photography discriminates between different types of "black". Some things black in visible light are "white" or "blue" in NIR light, but others are black or grey.

The third shot shows more of the same, but before reading on guess the colour of the woman's shirt.



Yes, you got it - "black"!. The man is wearing a black jacket and trousers and the woman has black shoes and a black pouch at her waist as well as the black shirt. The NIR light and camera response is signalling a range of difference.

The effect of lightening of a black coloured garment can be dramatic as the next image shows:



The lady is in a black trouser suit with shiny black shoes; the man in black jacket and trousers. Both are wearing white shirts.

Finally, just when I thought I was getting to know how the camera would respond I took this shot:



The lady was of Afro-Caribbean origin and, apart from gold pumps, was clothed entirely in black. Almost everything does turn out black in the image, except for the cape (if that is the correct description of that garment). This to my eyes is darkish blue/purple.

What is going on with all these variations in response? One way of explaining matters is to note that there are several types of dye used to create visible "black". One type is based on chemicals akin to carbon black (soot). This type of pigment absorbs more or less all visible and NIR light. Thus materials dyed with that type of dye will be black in the NIR image. Other dyes are mixtures of red, green and blue dyes that mimic black. Fabrics dyed with these substances may give a range of response depending on the NIR characteristics of the mixture. This is probably what is occurring in the blue/purple cape image. Can we say that this is why "black" shirts and suits become pale blue/white? possibly, but another explanation is that the nature of the fabric may be important - cotton, nylon, linen etc may have different NIR responses. But, that is for another time.

Cheers

David

[David]

In this last post on this thread, I want to look briefly at how different colours in the visible spectrum appear in the IR images of the converted Canon 20D. Remember that this applies to the R72 filter.

We know already that there is a wide and complex response from different colours: the green of foliage tends to light greys and white while black clothes can also appear white. I've offered explanations of why this occurs in terms of the spectral response curve for the sensors in the camera. However, can we find any general rules for predicting how colours will appear in an IR photograph? From reading around, most IR photographers would probably answer no. Nevertheless, having now taken scores of IR images I think some very general and rough rules of thumb can be discerned.

First, IR images ought to be appear as grey scale, but they do not because of the spectral response curve.

Second, foliage usually appears white or light grey, but may have red or blue casts, again from the sensor response to IR light.

Third, in non-foliage scenes (like urban landscapes) the yellow and orange region of the visible spectrum usually appears as white or very light grey. In fact, you could use the yellow lines on road markings to act as a white balance.

Fourth, in such settings, reds in the visible spectrum can vary from white through to dark grey in the IR image depending on the tonality of the visible red image.

Fifth, greens if they are pure green are the most likely to appear as greys, again depending on the tonality of the scene. However, many greens that are not foliage and that are found in the urban environment turn out to be blue-green in composition (even if the eye does not perceive this). Consequently, such greens can appear in the IR image with a blue cast.

Sixth, blues, as discussed in earlier posts, can be grey or blue.

Seventh, other colours are very difficult to predict.

Here are a couple of examples of colour versus IR images to illustrate these points. The colour shots were taken with a Canon Ixus 100 IS and the IR shots with the Canon 20D. The images of course do not line up exactly, but that is probably of no import.





In this first pair, we see that the yellows and reds of the billboard sign have been completely "whited out".

In the second pair below we see that while some colours change in a roughly predictable way others do not. - It all depends on the spectral characteristics of the pigment and how those characteristics match those of the spectral response curve.





There is potential for research in this area as there could be forensic uses for the IR versus visible response to colour in the environment. It may be that it is already used.

For those that want to further their IR photographic interests I suggest the following site:

http://r72.deviantart.com/

Many thanks for the interest shown in this thread.

Cheers

David

[Antonio Correia]

But Chris you are a master in this stuff !!!

This is a must read thread tomorrow morning

I knew IF existed but I had no interest but as I am trying new trends I want to dive in ...

[Dunkster74]

Hi All,
New to this site but fascinated by what I've read. Has anyone here combined an IR converted camera with neutral density filters in order to get long exposure IR images. If off to photograph the final space shuttle launch next week and thought I might try it. I'm looking to achieve a 2-4 minute exposure but don't want to loose the beautiful flame trail. Any advice or suggestions welcome!

Thanks

[Eric M]

Interesting thread. There is certainly a learning curve to IR. Mostly that entails using custom WB and editing technique. Having recently purchased a 665nm Canon G5, I was looking around the web for inspiration. This photog is my favorite for IR work. http://www.pbase.com/allonkira/infrared_photography

[Steaphany]

Be careful when photographing people wearing synthetic fabrics in IR. Where natural fibers, cotton, wool, hemp, tend to be opaque, some synthetic fibers are actually transparent and someone photographed clothed and expecting to be publicly modest may be imaged nearly or completely nude.

One of the early uses for night vision enhanced video cameras was to set it to "Night vision" mode while shooting through an IR filter to sneak nude videos of people in public.

This may be worth experimenting with if you have a cooperative subject who is fully aware of the potential nude images as they wear a variety of fabrics.

[JohnC]

These are wonderful ! I had to keep telling myself they were not taken in January. What an excellent thread.

[Eric M]

Hi Richard - Great shots - I've not had any focusing problems with the EF and EFS lenses, and I think that the firm that did the conversion for me take that into account somehow. As I mentioned in an addition above, I've got myself some Super Takumar lenses that seem not to show any hot-spotting at small apertures. The problem there, however, is that of manual focusing. In principal, then focusing for IR become a problem, but so far I have not had an issue with it in test shots. Further, those lenses have the magic IR mark on the aperture ring. But I'll have to check it all out when time permits.

Many thanks for your interest. I'm following on with another post on this thread about spectral response. I hope that will provide further information.

David

David,

By chance, have you used any of these lenses with the IR 20D? Sigma 10-20, Sigma 120-400OS, Canon 17-55 f/2.8 IS, or Canon 70-200 f/4L IS

It has occurred to me that my 40D would make a nice IR conversion but I had questions that a call to LifePixel didn't quite answer. Maybe you can help.