Register · Search · Software · Join Upload & Sell · Hosting

	Moderated by: Fred Miranda
	Username   Password FM Forum Rules FM Forums | General Gear-talk | Join Upload & Sell   Search Used end    Cutting haze with IR luminance detail

mpmendenhall Offline Upload & Sell: On p.1 #1 · Cutting haze with IR luminance detail I just tried this experiment in recovering distant/hazy detail by combining visible and IR exposure information, and thought it might be fun to share: Suppose you are shooting a distant subject on a not-perfectly-cold-and-clear day: No matter how good your lens/camera/tripod are, you'll likely be severely resolution limited because Rayleigh scattering by the air between you and the subject is scrambling your picture; in a 100% crop, there's little fine detail left: The strength of Rayleigh scattering is proportional to 1/wavelength^4 of light, so green/blue are much more strongly scrambled than red, or, even better, infrared. Let's take an IR picture of the same scene (B+W 093 filter; unfortunately, I don't have an IR-converted camera, so this takes a 30s exposure @ ISO200 in broad daylight, compared to 1/1000s @ ISO100 for the visible light): The infrared image "cuts through" the haze with much less scattering, so there is a lot more fine detail in a crop (plus, if all 3 types of color sensors on your camera are seeing the IR in fixed proportion, you can avoid the resolution-eating de-bayering color interpolation with a few tricks): IR photos can be nice on their own, but it's probably not always the effect that you want. So, why not take the best parts of the IR photo (the fine detail) and the color photo (the color) and combine them? Converting the color photo to a format where the luminance can be independently manipulated from the color (e.g. CIE L*a*b* color space), we can "slice off" the fine luminance detail from the IR photo (by subtracting off a blurred version of the image, e.g. by a 1 pixel gaussian blur), and add it to the color picture's luminance channel. The result combines the fine detail of the IR photo with the colors of the visible, revealing much more subject detail than is typically visible through the air: You can see some issues with this approach. One is the halo along the mountain edge --- which is exacerbated by the IR and color photos not being perfectly aligned (I did use the "align_image_stack" utility that comes with Hugin to attempt better alignment, but had trouble finding sufficient matching control points between the two images). Another is that the luminance noise of the long-exposure, higher ISO (+additionally sharpened) IR image is added to the color version. Both issues could be significantly mitigated in "real" images by using an image editor to blend selected portions of the detail-enhanced image with the color original (keeping fine detail on the mountain slopes, but erasing edge halos and sky noise). Although introducing some image artifacts, I think one can recover far more image detail this way that would be inaccessible by any other method. Feb 04, 2012 at 12:01 AM

dwerther Offline Upload & Sell: Off p.1 #2 · Cutting haze with IR luminance detail I have one word for you.... Decaf Feb 04, 2012 at 12:12 AM

Mike K Offline Upload & Sell: Off p.1 #3 · Cutting haze with IR luminance detail Another issue I have been struggling with for IR converted cameras, but its even worse in lens IR filtered cameras. CA. The specific refractive index varies with wavelength, and this gives rise to CA in normal visible photography. That is the different colors may not focus the same place on the sensor due to differing refraction with the various lens elements. Lens designers obviously have strategies for dealing with this phenomenon. Capturing both IR and Vis wavelengths and combining the result will result in exacerbating the CA since adding the IR is adding another broad spectrum range outside of the visible; the refraction differences will be larger than Vis alone. I looked on some of my lenses which have red markson the barrels for estimating the offset for IR infinity focus for IR film use. On some lenses, like my 24-70 f2.8L the red marks are really quite a distance from the infinity mark for visible light and vary considerably with zoom position. This means for this lens IR focusing has been calculated (and/or measured) by Canon to be significantly different for IR vs Vis and varies with variation in zoom. Could this the source of your alignment problems above? For IR modified bodies, there is a choice of different filters which let in increasing amounts of visible on top of the IR wavelengths. This has some benefit of creating more interesting false color images, but again, the wider the spectrum the greater the occurance and magnitude of CA (Vis vs IR CA). Some IR converter firms do note this as a disadvantage to using the wider bandwidth IR filters. Mike K Feb 04, 2012 at 06:08 AM

Mike K Offline Upload & Sell: Off p.1 #4 · Cutting haze with IR luminance detail The reason for less scatter of the IR in hazy conditions, from Wikipedia with selected edits for brevity: "Rayleigh scattering is the elastic scattering of light by molecules and particulate matter much smaller than the wavelength of the incident light. Rayleigh scattering... is inversely proportional to the fourth power of the wavelength of light, which means that the shorter wavelength in visible white light (violet and blue) are scattered stronger than the longer wavelengths toward the red end of the visible spectrum. This type of scattering is therefore responsible for the blue color of the sky during the day and the orange colors during sunrise and sunset." Mike K Feb 04, 2012 at 06:33 AM

Roland W Offline Upload & Sell: Off p.1 #5 · Cutting haze with IR luminance detail You should have added to your experiment the results of shooting the same scene with a high quality polarizer adjusted to show maximum general and color contrast. It will depend some on the angle of the light reaching the haze, but I find that distant haze is minimized in most shots, often by dramatic amounts. And of course the resulting image can be post processed for chromatic aberations to further help things along. As mentioned above, UV is the wavelength that is scattered the most. The classic generalization is that no UV filter is needed on modern digital cameras to minimize haze effects becasue the sensor has enough protection to block UV. But I guess I have never actually done testing to prove that is the case with my particular cameras. If I get a chance to do so some comparisons some time soon, I plan to include both a good UV filter and a good polarizer in my mix, as well as combining both types of filtering. I also now wonder if there might be different grades of UV filters, and that some would cut off sharper and better, even at the cost of slightly cutting in to the blue end of visible light. Such a "special" UV filter might help even more with minimizing haze effects. Feb 04, 2012 at 08:11 AM

mpmendenhall Offline Upload & Sell: On p.1 #6 · Cutting haze with IR luminance detail Mike K wrote: Another issue I have been struggling with for IR converted cameras, but its even worse in lens IR filtered cameras. ... Could this the source of your alignment problems above? You are correct that CA can be a serious problem in many cases for imaging outside the visible. However, for this test I used a lens specifically designed to minimize color aberrations from 400 to 1000nm wavelengths (Zeiss 250/5.6 Superachromat). The alignment difficulty in this case comes from the fact that it's virtually impossible to keep a longer lens aligned to pixel precision while you are swapping out filters. With a sufficiently wide angle lens and sturdy tripod, this might not be a problem (though CA might be). The alignment program I use could help correct for lateral CA (though not missed focus due to focus shift in the IR) by re-sizing the IR image to match the field of view of the color (much like visible light CA correction tools can re-size the color channels to align). Roland W wrote: You should have added to your experiment the results of shooting the same scene with a high quality polarizer adjusted to show maximum general and color contrast. It will depend some on the angle of the light reaching the haze, but I find that distant haze is minimized in most shots, often by dramatic amounts. And of course the resulting image can be post processed for chromatic aberations to further help things along. As mentioned above, UV is the wavelength that is scattered the most. The classic generalization is that no UV filter is needed on modern digital cameras to minimize haze effects becasue the sensor has enough protection to block UV. But I guess I have never actually done testing to prove that is the case with my particular cameras. If I get a chance to do so some comparisons some time soon, I plan to include both a good UV filter and a good polarizer in my mix, as well as combining both types of filtering. I also now wonder if there might be different grades of UV filters, and that some would cut off sharper and better, even at the cost of slightly cutting in to the blue end of visible light. Such a "special" UV filter might help even more with minimizing haze effects. I have tried UV filters, and found them not to help with situations like this. The loss of detail is not just due to a short-wavelength haze being added over the picture (which might be filtered out), but all wavelengths are getting muddled by Rayleigh scattering, including the whole green channel where the camera sees most of its luminance detail (two green pixels for every one blue and red). A polarizer would have made an interesting comparison (based on the physical cause of the blurring, I doubt it would help much with resolution, but can improve color clarity). Feb 04, 2012 at 03:44 PM

jcolwell Offline Dedicated FM Upload & Sell: On p.1 #7 · Cutting haze with IR luminance detail Here's a quick comparison of some configurations I tested today. 1DsIII + 100-400L IS at 400mm f/5.6, ISO 400, handheld, IS on. 1. top left UV filter, yes hood (B+W MRC UV-haze 010M) 2. top right, no filter, yes hood 3. bottom left, CP filter, no hood (B+W MRC KSM) 3. bottom right, CP filter, yes hood (B+W MRC KSM) The first image shows the full scene, followed by four 100% crops of the lighthouse and then four 100% crops of the navigation buoy. Each image is the best of three that were shot for each combination. The light is very grey and diffuse, from high overcast clouds. The harbour water is about 4 degrees warmer than the air. I think 'no filter' provides the best detail, CP filter provides the best contrast and discrimination between similar tonal areas, and the UV filter is relatively poor. YMMV. Feb 04, 2012 at 08:01 PM

Ruahrc Offline Upload & Sell: Off p.1 #8 · Cutting haze with IR luminance detail Brilliant post! A very unique method to enhance image quality. I wonder if someone were to repeat this with the Coastal Optics 60mm UV-VIS-IR lens, what the results would look like (The Coastal Optics lens is specially made to operate in all wavelength regimes, and from what I hear is among the best corrected lenses available). It was always my understanding that virtually all meaningful amounts of UV wavelengths are absorbed by the glass elements in the lens anyway, so adding a UV filter on top doesn't really do anything extra. That's why pretty much any UV instrument uses quartz instead of glass. Norman Feb 05, 2012 at 10:49 PM

vachss Offline Upload & Sell: Off p.1 #9 · Cutting haze with IR luminance detail This is related to an idea I had to reduce diffraction effects in extreme macro by isolating the blue channel, and converting to luminance. In principle should give almost twice the resolution, but the problem (as in the OP's example above) is that luminance info derived from a single wavelength band doesn't really replicate the RGB result and trying to recover a color image is problematic. Feb 06, 2012 at 04:39 PM

HerbChong Offline Dedicated FM Upload & Sell: On p.1 #10 · Cutting haze with IR luminance detail actually, the lens is near enough opaque to UV that it really doesn't matter much what the sensor does. the fewer the elements, the more leakage of UV but even with a 4 or 5 element prime, it's pretty minimal. this is from shooting with a Fuji IS-Pro which is sensitive to about 350nm in the near UV. Herb... Roland W wrote: As mentioned above, UV is the wavelength that is scattered the most. The classic generalization is that no UV filter is needed on modern digital cameras to minimize haze effects becasue the sensor has enough protection to block UV. Feb 06, 2012 at 06:11 PM

mpmendenhall Offline Upload & Sell: On p.1 #11 · Cutting haze with IR luminance detail vachss wrote: This is related to an idea I had to reduce diffraction effects in extreme macro by isolating the blue channel, and converting to luminance. In principle should give almost twice the resolution, but the problem (as in the OP's example above) is that luminance info derived from a single wavelength band doesn't really replicate the RGB result and trying to recover a color image is problematic. The other problem with isolating the blue channel is that, in a typical camera color filter array, only one in four pixels records the blue channel (another 2 in 4 green, 1 in 4 red). The rest of the resolution in the blue channel is interpolated with "help" from the other color channels, thus is at best no higher in resolution than the more diffracted red and green channels. For my IR image above, I used a B+W 093 filter (=Schott RG830, 830nm long pass) which blocks out virtually all the visible light, after which all three types of color sensor in the camera have similar amounts of "leakage" in the IR. This way, I was able to extract luminance detail using the full resolution of the sensor, instead of a coarser single color channel. I think that, for increasing macro resolution, focusing stacking is the most promising technique, since the "slices" of the image can be taken at a wider aperture with less impact from diffraction. Feb 06, 2012 at 11:03 PM

	FM Forums | General Gear-talk | Join Upload & Sell end

You are not logged in. Login or Register

Username   Password   Lost your password?