Hi everyone,
I am using Fiji to determine the percentage of the image covered by leaves, branches, or trunk, but i've encountered some difficulty when the sun hits parts of my image like the one below. Is there any way I can still measure the canopy cover but include the trunk, even though it is bright like the sky? I've tried colour thresholding but I haven't been able to make it work.
Thank you in advance for your help and advice,
Mondane
I do not use FiJi / ImageJ much so I'm sorry I can't direct you more precisely.
I've tried (very quickly) the following, which is available using FiJi :
- converting the RGB image to a Lab image : I thought this might contrast your sky better vs. trunk, branches and leaves, which it did.
- extract the b layer (that's the yellow/blue axis in the CIE Lab colour space)
- manually segment the sky (tried out various automated thresholding algorithms, not much success).
It worked better than I imagined this quickly, but still some work to be done. You might start with this to get a quick estimation, but I'm pretty sure it would not work all the time (cloudy sky, night sky ?).
Deep/Machine learning thresholding techniques may very probably work better and be more robust, but I'm not an expert on those, especially using FiJi.
I tried Pierre's technique and it works well.
I found that the issue with thresholding your image is less about the sky and more about the tree trunk. Both thresholding and color thresholding select too much of it to give you an actual value of leaf area in your image. You need a way to pull out or separate the colors you want from your image. Therefore, I would also suggest trying the colour deconvolution plugin (https://blog.bham.ac.uk/intellimic/g-landini-software/colour-deconvolution-2/). It is a plugin that I have found to be very useful.
You will want to use the From ROI option, using the rectangle drawing tool to select desired and undesired colors in your image when the plugin prompts you to do so. As a test run, I selected leaves, the trunk's medium grey, and the trunk's dark grey. When the three separated images come up, close all but the one for the leaves (in my example "Colour_1"). The attached image is the result. It is pretty good, but you can play with the color values for the ROIs to fine tune the selection a bit. If you have multiple images with the same features and color properties, you can turn your ROI matrix into its own vector (follow the instructions on the plugin's website linked above) to save a little time. Then threshold and quantify the image as normal.
Here are the values that produced the attached image:
From ROI Vector Matrix ---
Colour[1]:
R1: 65.25852
G1: 56.522984
B1: 144.18433
Colour[2]:
R2: 62.1708
G2: 63.064983
B2: 65.657684
Colour[3]:
R3: 96.13901
G3: 100.14818
B3: 110.26313
From ROI Java code ---
if (myStain.equals("New_Stain")){
// This is the New_Stain
MODx[0]=0.38831538;
MODy[0]=0.3363353;
MODz[0]=0.8579567;
MODx[1]=0.5639476;
MODy[1]=0.5720587;
MODz[1]=0.595577;
MODx[2]=0.542282;
MODy[2]=0.56489617;
MODz[2]=0.62195057;
I hope that helps. Good luck!
- Melissa
- Badges
- Science topic
- Similar topics
- Computer Science
- Images