Hi all, I was wondering if anyone knows the underlying calculation or equation used to get Tamura Granularity in the image features node? I can’t find the literature that describes this exactly.
My issue is that it takes quite a long time to calculate the granularity but this feature works really well to estimate thresholds in my datasets. I was hoping to understand how it is calculated to see if there is anyway to speed up this process.
Hi @aseeber
The seminal paper by Tamura about textural features published in IEEE is the following:
After checking the -Image Feature- node, I notice that it uses all the processor of the computer, in my case 12, but it doesn’t fully use them 100% as shown in the snapshot below:
If you are running this node in a computer with several processors, I would recommend to use the -Parallel Chunk Start- and -Parallel Chunk End- nodes, to increase the use of parallelism, as shown below when using them with the Image Features nodes inside the parallel loop:
As a result, one can see that the processors are now fully employed leading eventually to a gain of time if the node is run in this way, as shown below:
Hope this hint helps.
Best,
Ael
For completeness, the actual implementation is here:
@aseeber thank you for your feedback and for validating the solution.
Best,
Ael
Not sure where to report node bugs, but we noticed while rewriting this that line 100 should be:
border = Math.pow(2, k);
Instead of:
border = Math.pow(2, 2 * k);
Also, at line 170, we replaced 3 with 6 since this is maximum k.
So the first point is a bug fix and the second is a suggestion. @imagejan point in the right direction for bug reporting if you don’t mind.
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.