At this point I haven't put much thought into ideal targets mostly because I don't have access to a spectrometer to measure and test different targets. I expect some commercial folks are working on it so maybe they'll come out with something that isn't too expensive. Most of the professional calibration material if very pricey.

QR code libraries are available in many languages, and most QR code libraries will probably return a location of the center of the code. But maybe not an orientation. So I figure repeating the QR code with samples along the axis at known distances (say, every 20% of the way between the two targets) is an easy standard to stick to, with some futureproofing. How does that sound? Any preferred ordering of the patches?

I'll probably do a version in JavaScript.

The panel order is not important as long as it's predictable and known. The "ideal" targets would probably be a few gray patches from bright to dark and it would be nice if the refelctance of each target was roughly similar through the visible and NIR. Some other qualities are that the samples should have diffuse (as opposed to specular) reflectance properties and as you noted the reflectance properties should be reasonably stable so they don't change over time. Weatherproof and protected would be useful for working in the field. When sampling form the calibration patches you'd make sure you sampling pixels away form the patch edges. I'm sure there are other qualities/features that other folks can add to but this is a start.

This is something that'd also benefit the general white balance process of Infragram cameras. I wonder, are there any paint sample strips that are really long-term stable, and widely available? I think we asked this early in the Infragram project too.

Have you considered using reference charts used to calibrate green rooms in cinema? Some thing like this: labsphere - Reflectance Reference Targets Tetra cam used Multi-spectral Ground Calibration Targets

Rosco manufactures outdoor-grade TV paints including 60% grey (TV white) and black (3%) and chroma colors (blue, green). They would be good choices for a DIY calibration card. https://www.rosco.com/FTVP/TVPaints.cfm https://www.rosco.com/FTVP/ultimatte.cfm

Mathew - Have you been able to find a reflectance plot for those paints? I looked but didn't see it. Other paints I was looking at a couple years ago had plots but only for the visible range. It's important to know the spectral reflectance properties into the NIR and having some assurance that it's stable over time and consistent from batch to batch would be nice too.

Please help. If it is possible to describe all the work with the plugin. How to create "сomma-separated variable (CSV) file with the reflectance values for each of the reference targets in the image that you want to use to calculate the calibration coefficients." I am using mobius cam from http://infragram.org/. Thanks!

Hi aldehyde - Did you look at the user guide on Github: https://github.com/nedhorning/PhotoMonitoringPlugin that should have information about creating the CSV file of target reflectance. Here is the excerpt from the user guide:

Before running the plugin you must create a comma-separated variable (CSV) file with the reflectance values for each of the reference targets in the image that you want to use to calculate the calibration coefficients. The rows of the CSV file correspond to the targets and the columns are the target reflectance values for the visible and near-infrared wavelengths that correspond with the camera filter being used. For example, the following values are for two targets (first line is the bright target and second line is the dark target): 0.86696300, 0.90032700 0.04748605, 0.05665055

How to get target reflectance values for the visible and near-infrared wavelengths that correspond with the camera filter? Can I use?. 0.7529225,0.860895 0.753183,0.751541 0.854773 0.900327 0.0418333,0.0427029 0.3103265,0.3549215 0.04686155,0.05665055 (Red card stock - leftmost target White spray paint White printer paper Black spray paint Gray spray paint - rightmost target Tar paper - off on its own)

Unfortunately those values would only work if you had the same materials that I used. To get the reflectance values you would either need to fine someone with a spectrometer who could scan your targets and give you the reflectance values for the wavelengths suitable for the filter you are using or purchase targets that have documented reflectance values which is usually quite expensive. Getting calibration targets that have known reflectance values seems to be the biggest hurdle for the calibration process. Most universities have spectrometers (lab and or field) so maybe if you ask around you can find someone to help.

Hi Ned,

I am really excited about the work you are doing here as it helps resolve some major issues with my grad studies. I have a few questions:

• When I view the jpeg EXIF in Irfanview, I see "max aperture" but not simply "aperture". Is there a way to determine if my images have identical aperture?
• I am looking at this as a method to resolve our lack of RAW images. I am planning to use your workflow, then use the calibration function of Pix4D when generating a reflectance map. Is this a good idea? In the future, I will shoot RAW, convert to Tif, and let Pix4D do its thing.
• How many pixels are needed to calibrate? Since we do not have pre-flight images of the target, we are stuck with a roughly 20x20 pixel ROI. *Would you be able to post your R scripts for gamma correction?

Thanks for all the effort you put into this. It's great to see opensource processing receive features that proprietary software seems to lack.

Hi Jacob, To check for aperture you could try another EXIF reader. Maybe "max aperture" is the same as "aperture" but I have no idea. When working with Pix4D if you are planning on stitching NDVI images that might be problematic since it is more difficult for Pix4D to locate unique keypoints to match in adjacent images. I would like to add a feature to the photo monitoring plugin to output calibrated image bands but am not sure when I'll have time to do that. Using calibrated image bands would probably be the best approach for creating mosaicked images (RGB and NDVI) in Pix4D and other similar software. A 20x20 pixel ROI should be fine to calculate calibration coefficients as long as the ROIs are fairly homogeneous and without edge effects where the edges values are different from the rest of the ROI.

Thanks for the feedback.

I think I misunderstood the function of this plugin. I thought I could calibrate my jpeg R/G/NIR images to a black, a grey, and a white panel of known reflectance, while attempting to return some of the linear-response of a RAW. NDVI would be done separately. My proposed workflow is:

• Pre-process vignetting, dark current, etc. in Lightroom or DCRaw using custom lens profiles
• Run those outputs through the plugin to calibrate to different shaded targets
• Orthomosaic and generate reflectance map in Pix4D
• Calculate NDVI and other VIs in QGIS using raster calculator

The current version of the plugin only outputs index (NDVI and DVI) images. The plugin only calibrates two bands, a visible and a NIR band and two output a reflectance image I would with have to output two single band images or a typical 3-band image but that would require that I decide on how to deal with the third (often the green channel) band. A few people have requested this feature so I'd like to do it but not sure when I will happen. Perhaps in January.

Hello, @nedhorning, I had recently been researching about NDVI and came across your amazing calibration plugin. It works great; however I have a few question for you, if you don't mind answering them.

I am doing a research on corals in a fish tank right now and I came across with these problems

1. Does the reflection of the glasses effect the result of the photo as you might noticed the light from the lamp is extremely bright compare to any other objects in the photo.
2. I tried calibrating with the CSV file you indicated in the article; however it seemed to have failed
   • White paper 0.866963, 0.900327
   • Black magnet 0.047486, 0.056651
3. Also, will water effect the image too?

As you can see, the light that is reflecting has the highest "NDVI" value and also the reflection from the mirror also shares exact NDVI value as the object.

Thank you very much and I am sorry if I asked some stupid question because I know very little about photo shooting.

And the reason why i flipped the min max value is because it just looks more obvious that way, so I am not sure what effect it causes to the image.

I used Rosco Roscolux #2007 Storaro Blue Gel Filter

Top_Color.tif

Hi David - This is an interesting project. The reflection of the glass should affect the results as long as the reflection is not in the area of your calibration targets. If there is glass between the camera and the scene you are photographing that could have some impact on the results since the quality of the light might change as it passes through the glass. Water will have an effect on any near-infrared (NIR) imaging since most NIR light is absorbed by the water.

Another possible problem is that you use the reflectance values for the calibration target I used and they might not be accurate for the targets you are using. It's possible that the values are similar but I don't know for sure.

You didn't mention what type of filter you are using. From your settings screenshot it appears as if it is a blue filter. Is that what you are using? Another (perhaps the main) issue is that the settings screenshot is from the plugin that does not calculate or process the calibration information. For the calibration you need to use the "Calculate image calibration coefficients" plugin. Let me know if that helps.

David,
When using a blue (#2007) filter, the red channel records NIR and the blue channel records visible blue light which is used to compute NDVI. Water absorbs NIR two orders of magnitude more strongly than it absorbs blue light. It will be very difficult to capture a photograph which has good information about both blue and NIR because one of those channels will typically be either overexposed or underexposed.

Light absorption coefficient of pure water. The absorption coefficient (y axis) is a log scale so NIR light (e.g. 760 nm) is absorbed 10 times faster than orange light (625 nm).

Below is the color histogram of the bluish photo you included in your note. The blue channel (which records visible blue light) is completely overexposed (pegged at ~255). All the pixels in the photo essentially have the same value for blue, and that value is used to compute all the NDVI values.

Histogram of the bluish photo in the note. The blue channel is severely overexposed.

It should be possible to take a photo with less overexposure in the blue channel and not too much under-exposure in the red (NIR) channel. It will help to shoot through as little water as possible. It might not be possible to get NDVI results similar to those from plants in air, but your results could still have meaning when compared with themselves.

Chris

Thank you very much for the advice Chris and Ned, I have learned a lot. I will continue to this research, so if possible if I encounter any problem, can I leave my question here?

Fyi: I have tried calibrating the image; however it shows an error of Error: java.lang.NumberFormatException: For input string: ".3.5"

Hello, @nedhorning, for the Rosco Roscolux #2007 Storaro Blue Gel Filter, with the following datasheet

Could I just use the visible reflectance at about 440nm, and NIR at about 740 nm?

Are a cardboard, a white paper and a black Ethylene-vinyl acetate (EVA) good targets?

Hi - Using 440nm seem reasonable but I'd probably increase the NIR value to 760nm or beyond or something like that since the peak is probably beyond 740nm. That's just a guess since the graph is cut off. As far as advice for selecting calibration targets you could test the material you suggest - assuming you have access to a spectrometer. The best targets will have relatively flat reflectance curves (reflectance for all wavelengths are similar) but that can be difficult. In your case it would be best if the reflectance curve was flat in the NIR wavelengths since we don't know where the peak is. It's also good if the calibration target is not very sensitive to specular reflection. It's better if the light is reflected diffusely - equally in all direction.

Dear Ned Horning,

I calculated image calibration coefficients, using the reflectance data from department's spectrometer and 3 calibration targets. I found a good correlation of R2 0.9432 for Visible which at 600 nm and R2 0.84332 at reference 850 nm for NIR. But looking at the graph , i was wondering what the image values with the range (0.0-1.0) ? How i can calculate the image values manually using ImageJ and shouldn't this value be something between (0-255) as i saw in some other posts at public labs?

I look forward to hear from you Ned.

Thank you so much

Hi Suman - The values from 0 - 1 have been scaled from the original image. The reason I do this is that some image pixel values range from 0 - 255 but others can be between 0 - 1024 or 0 - 2048 for example. If you wanted to scale an image from 0 - 255 to 0 - 1 you could use this formula: Pixel_value * 1/255 Another point is that if you only have two targets then the R2 will always be 1 or very close to 1 even if the actual fit between pixel values and ground reluctance is not very strong.

Ohh I get it now. Actually the data has 3 calibration target, but i plan to increase it to 5 targets. I also wanted to know, how i can manually measure the pixel value through a rectangle selection in ImageJ. I tried using Analyze --> Measure, but all the calibration target that i measured through individual rectangle selection has similar range of mean values (250-255 range)

Thank you so much for your prompt response Ned...

Hi Suman - Using Analyze => Measure should work. It's not uncommon to have high pixel values for bright targets. In fact if the exposure isn't set well the brighter targets are easily saturated at 255. If you have calibration target images with a lot of pixels at 255 they have almost certainly saturated the sensor and it would be best to retake those photos with a lower exposure value to create a darker image using a faster shutter speed, smaller aperture or lower ISO speed.

Hello Ned, I tried using different exposure settings now, and its working, the pixels aren't saturated anymore.

Thank you so much for your prompt response as always. Highly appreciable

Best Regards,

Suman