Experimenting with NDVI , fiji + photoscan

Hello All,

I just started experimenting with NDVI. I bought a IR-PRO.com INFRABLU NDVI LENS for my HERO3 black edition.

This is what I did:

- I went and did a small survey mission at a field and processed the orthomosaic in photoscan.

- I then opened up the ortho in FIJI Image and processed the NDVI.

Using NDVI_VGYRM.lut color map from http://publiclab.org/notes/cfastie/08-26-2014/new-ndvi-colormap

I noticed one thing. If I process the NDVI for a single image that made up the orthomosaic, I am getting mixed colors vs the orthomosaic which is just green mostly.

I have attached both images.

Let me know what you think.



You need to be a member of diydrones to add comments!

Join diydrones

Email me when people reply –


    • Ok great.
  • Photoscan may or may not be resampling the images (changing the resolution). To make that nice seamless mosaic, they also have to alter the color or brightness of images to make them match each other. The dynamic range of the final mosaic might be similar to the range in some of the original photos, but to accomplish that, the dynamic range of many individual photos might have to be reduced. That could compress the NDVI values represented in each individual image. That is what you observed for the single NDVI image you posted.

    Another similar possibility is that to make the individual photos match each other better, Photoscan alters the color of the photos. That could make the values in the red and blue channels more similar. If R and B are more similar, R minus B will be smaller and computed NDVI values will be lower.

    Maybe Photoscan has an option to preserve the color and brightness of the original photos. The final stitched mosaic would look terrible (because the individual photos would match poorly), but it could provide a hint about why the current mosaic has such compressed NDVI values.

    Although your current mosaic seems to have less information than the individual photos, maybe what you have lost is mostly the outlier NDVI values. All the red and magenta in the individual NDVI image is very high NDVI which is probably just artifacts of deep shade (a common problem). The lack of those artifacts in the final mosaic may be a blessing in disguise. Your current NDVI mosaic looks like an excellent result. 


    • Thanks a lot!

      I will check if photoscan has any such option.

  • What is the normal process for aerial NDVI? Are people doing orthomosaics from aligned images or just super high altitude with 1 image ?

    • Most folks are stitching multiple images for any significantly-sized field.  It really depends on the resolution you need.

      How many photos make up your orthophoto?  If the original image is 4000x3000, and the ortho is 7500 x 15000, it feels like your ortho is a pretty significant downsampling from the original resolution.  For an 8 acre plot we flew last year, our output image is 18000x15000 @ 2.5 cm ground resolution.

      When you export your orthophoto, you should have the option to specify pixel resolution in meters (?).  Try setting it to something like 1 or 2 cm and see if that changes anything.  Your problem could be from other things, but I definitely feel like resampling is potentially the fundamental problem, causing you to lose data.

    • Ok Kyle,

      I did that, output image is larger now 29175x2025. However same result in fiji.

  • Without having the original files, it's hard to say exactly what is happening, but basically it appears that you're losing fidelity in the stitched image so that all of the values end up being a little higher.  

    What resolution is your exported ortho versus the single image?  If you're exporting a low resolution ortho, then it's going to resample the data, and you're going to lose pixel values.  I imagine photoscan is doing some cubic resampling algorithm to downsample, and this may be dropping the values you see in your single image.  Try increasing the resolution of the ortho and seeing if that helps.

    • The resolution exported from photoscan is 7549 x 15000.

      The single images are 4000 x 3000

This reply was deleted.