Public Lab Research note


Testing MIDOPT filters

by cfastie | December 03, 2015 03:54 | 2,184 views | 8 comments | #12473 | 2,184 views | 8 comments | #12473 03 Dec 03:54

Read more: stable.publiclab.org/n/12473


Above: A folder full of photos of the same scene taken with different filters and different white balance settings.

Last spring, Barry Warzak of Midwest Optical Systems (MIDOPT) in Illinois contacted Ned Horning about writing an article on dual bandpass filters. Ned got me involved, and Barry also became interested in having us try some of the new filters he was developing. The primary focus of MIDOPT is industrial machine vision filters and lenses, but Barry recognizes the potential in the emerging market for agricultural analysis and aerial imaging. Barry has sent us samples of 15 different filters that could be used for infrared plant health analysis. Some of these are fancy dichroic filters that transmit one or two very narrow bands of color, and others are high quality glass filters that have spectral qualities similar to Wratten 25A, Wratten 15, or Schott BG3 filters. We have posted a few research notes about our use of these filters, but there is much more we could learn from them.

Tube-703-7.jpg Tube-703-9.jpg
Above: I taped a tube for 52 mm threaded filters to a PowerShot A810. But the shiny ring around the lens reflected off the inside of the filter and ruined some of the photos.

Ned has been using the filters in his experiments with calibrating NDVI images, and Barry recently sent us samples of calibration targets of known spectral reflectance. This type of target is placed in the scene to be photographed so the brightness of each pixel in a photo can be related to actual reflectance or radiance at different wavelengths. This is required if the NDVI values derived from photos are to have values that can be related to legacy NDVI (e.g., from satellites).

Tube-703-13.jpg Tube-703-21.jpg
Above: We wanted camera RAW image files, so I removed the IR block filter from a PowerShot S110 which can save native raw files. Reflections from the silver lens housing would reflect off the filter, so I taped some black felt to the front of the lens.

To use the filters we removed the IR block filter from inside some Canon PowerShots and placed the filters in front of the lens. Filter tubes for certain PowerShot models can be taped onto the front of any PowerShot and allow the filters to be easily swapped. Our standard protocol is to set up a solid tripod over a scene with some green plants, dead foliage, and calibration surfaces of known reflectivity. Photos of the same scene can then be taken with a series of different filters.

Midopt-700-7.jpg
Above: Filters are easily swapped so the same scene can be photographed with lots of different filters.

Midopt-700.jpg
Above: The filter case with the printed filter name was put in the scene where it could be read in the photo (except when I forgot). I also kept a written list to record white balance settings and other notes. The S110 saved jpegs and camera raw files for each photo.

IMG_0091.jpg
Above: For this photo of the test scene, an IR block filter was put in front of the lens to replace the internal one I removed. The four dark squares on the pine board are targets with different reflectances. Each target reflects a particular percentage of light regardless of color, from 350 nm to 1200 nm.

Ned will use these photos to refine his calibration routine. Below are some of my attempts to make NDVI from jpegs without calibration. So although the NDVI images distinguish healthy foliage from non-foliage, the particular NDVI values represented are not correct except by chance and brute force (from the white balance procedure).

IMG_0109_660-850CWB2.jpg
Above: This bluish photo was taken with a dual bandpass dichroic filter whose spectral transmission curve is below. The blue color is due to a custom white balance procedure done to the camera to make the blue channel brighter. The blue channel captures mostly near infrared (NIR) light. Note the vignetting at the corners.

660850peaks.JPG
Above: The transmission curve for this filter. Only red light and NIR light can pass through the filter. There is more about how these filters are used in this note.

IMG_0109_660-850CWB2split.jpg
Above: The three channels in the bluish photo above. The red channel (top) has captured mostly red light, but the green (middle) and blue (bottom) channels captured mostly NIR light. The blue channel has slightly more vignetting (dark corners), so I used the green channel for the NIR values when NDVI was computed.

IMG_0109_660-850CWB2bothR_G.jpg
Above: NDVI derived from the red and green channels of the photo above. There is good discrimination between healthy foliage and non-foliage. The brown tree leaves might still have pigments that are absorbing red light and reflecting NIR.

NDVI_VGYRM_lutA.jpg
Above: The color scale applied to the NDVI images above and below. Healthy foliage typically has NDVI values between 0.2 and 0.7.

IMG_0116_650-808CWB2.jpg
Above: This photo was taken with a dual bandpass dichroic filter similar to the one above, but the transmission peaks are shifted slightly (to the left). The red band is centered on 650 nm (instead of 660 nm) and the NIR band is centered on 808 nm (instead of 850 nm). The same white balance setting was used. Note the stronger vignetting.

IMG_0116_650-808CWB2Split.jpg
Above: The three color channels in the photo above. The blue channel (lower) is strongly vignetted. Although the green channel (middle) also captured mostly NIR light (note the similar differences between foliage and non-foliage) the green channel is not vignetted. I don't understand why there is this difference.

IMG_0116_650-808CWB2both.jpg
Above: NDVI image using the red channel (red light) and blue channel (NIR) from the photo above. The dramatic vignetting is enhanced by the particular color table applied (same table as for the previous photo) but with any color table the vignetting would be distracting.

IMG_0116_650-808CWB2bothR_G_12.jpg
Above: NDVI image using the red channel (red light) and green channel (NIR) from the photo above. Vignetting is mostly absent. This result is very similar to the NDVI result from the other filter (above).

The dramatic difference in vignetting between the two filters is unexplained, as is the difference between the green and blue channels in the photo with the second filter. What observations could I make which might reveal why these differences exist? Sweet mysteries.


8 Comments

@cfastie -Hi Chris, Is it possible that the vignetting your seeing is related to wavelength interference due to aperture size? Have you tried a larger aperture to see if the vignetting is still so pronounced?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


I hadn't considered aperture as a reason for the vignetting. I assume it is related to the angle at which the light enters the filter. Dichroic filters are intended to work only for light impacting the filter at an angle close to perpendicular to the filter surface. The camera is capturing light that has passed through the filter at much lower angles, and that light is not "filtered" the same as more perpendicular rays. This effect seems to be much stronger for near infrared light than for visible light, and stronger for 808 nm NIR than for 850 nm NIR.

A solution might be to hold the filter closer to the front of the lens, or install it inside the camera in front of the sensor. It might also be helpful to zoom the lens in, or just use the center part of each photo.

The mystery of why the blue channel showed the vignetting more strongly than the green channel remains. Maybe the green channel captures a different range of NIR light than the blue channel. The range of NIR passed by the filter is very narrow, but that is only for light passing through the filter perpendicularly. A wider range of NIR probably passes through the filter at lower angles, and the Bayer filters for the blue and green channels might block different parts of that range.

Reply to this comment...


@cfastie -I can confirm that mounting the filter internally does correct the vignetting problem. I have mounted a DB475/850 and DB660/850 in place of the IR filter on Canon S100's and there is no vignetting.

Reply to this comment...


Hi @Claytonb,

That's good to know. Did you compare the different channels to see if some showed any vignetting? A photo of an evenly lit solid field would be a rigorous test of this.

Do you know the thickness of the filters you installed internally? Have you examined the image quality (crispness) from edge to edge with the new filters in place? I have had trouble getting good focus when a glass filter 0.60 mm thick was installed internally (see this note). I tested focus at the edges in photos of a subject far from the camera to mimic aerial photos.

Chris

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Wow, the 660/850 camera focuses well across the image. That's good to know it can be done with a 1.0 mm thick filter. The 475/850 camera looks a little funny at the edges, but maybe that's the filter doing something weird with the wavelengths.

You are right that there is essentially no sign of vignetting. That's also good to see.

The exposure on those photos is perfect. In all three color channels, the white target is not overexposed (it's < 255) and the black target is not underexposed (it's > zero). That is not easy to do.

Are your NDVI images calibrated or based on the custom white balance?

Thanks for posting these.

Chris

Yikes, where did your comment go?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


@cfastie Apologies Chris I had to fix the image layout and labels and couldn't edit right so I'm reposting.These are from the MidOpt DB 475/850 and 660/850 filters.These are uncalibrated and were converted from RAW. All taken with f/4. exp 1/1000, ISO 100. The 475/850 used the camera's cloudy white balance and the 660/850 was white balanced with red construction paper in the shade. 475/850 jpeg

475-850_jpeg.JPG

475/850 VIS Red NIR Blue 475-850_VIS_Red_NIR_Blue.jpg

475/850 VIS Red NIR Green

475-850_VIS_Red_NIR_Green.jpg

660/850 jpeg

660-850_jpeg.JPG

660/850 VIS Blue NIR Red

660-850_VIS_Blue_NIR_Red.jpg

660/850 VIS Green NIR Red

660-850_VIS_Green_NIR_Red.jpg

Reply to this comment...


The filter size I have in the Canon S100's are 8.9 X 7.9 X 1. The sensor screws are each backed out approximately 3/4 turn from the factory position after placing the filter and then a lot of small adjustments to get the focus close.

Reply to this comment...


Okay, looks good. You just have to put carriage returns or two spaces after a line or an image to make it insert a return.

Reply to this comment...


Login to comment.