I have two sets of canopy pictures from a forest taken at dbh with a fisheye lens. One set of pictures are from the spring just before the trees get their leaves, the other is from late summer where the trees have all their leaves. Both times the pictures where taken in the same spots.
I´m have been analyzing these two sets of pictures separately in R in order to get a value of the effective leaf area index from each picture to use as a proxy for the amount of light available for the flora. The aim is to be able to link the amount of light available on the forest floor to the floral development.
But I have encountered a problem when choosing a method for setting a threshold and thereby binarizing the fisheye picture to further analyze it. For the pictures taken in the summer I´ve used the Otsu method, since it seem to work well here. But when analyzing the spring pictures this method doesn´t seem to work well, since it´s not cooping well with the clouds in my pictures. Instead I have found, that the Huang method works well here.
So my question is, can I compare my results when I have used two different thresholding methods?
And if not, which method would then be cooping best when I have pictures with both direct sunlight and dark clouds?
Dear Katrine,
So far as I know, all pictures should be taken under overcast sky conditions. Did you do so?
You can optimize the threshold using the free software SideLook 1.1 (Nobis 2005; Nobis, Hunziker 2005). I do not remember which algorithm uses the software, but it should solve your problem of using two different methods. With this software, you have some control of the threshold assignment and you can use several color channels (blue, green, red, and grey, respectively) that can help you improve differentiation between black and white pixels. The blue band is usually best, but you could try another band for sun vs cloudy images. You should have no problems if the black pixels represent only canopy elements. A visual inspection after thresholding should be done to see if clouds are not taken as black pixels. As a last resource, the old way was a visual inspection of every picture can be used, as thresholding was developed as a method to reduce time spent in this step.
Regards
- Badges
- Science topic