Coefficient of Variation for SAR images - CoVAmCoh method

Hi. I am desperately striving to reproduce a CoVAmCoh processing on my SAR images. It works with interferometry pairs and consists of a combination of 3 bands ==> Coefficient of Variation, a Mean amplitude and a Coherence image. The issue lies in the making of the CoV image. I use the “coef_var” function with the two intensity images from a coregistered interferometry stack. I get a result but doesn’t seem to be quite consistent with the thing I can get from the same function in ERDAS. Too much noise as it is computed on 1 pixel and not a definable window (3x3, 5x5, 7x7 …). Is there any specific parameter to give or something to add to make it run properly ? Any suggestion ? Thanks.

Does the documentation in ERDAS describe exactly what the corresponding function is doing?

Not exactly. It says some basic thing like “The Coefficient of Variation is the ratio of the Standard Deviation and the Mean, in a local neighbourhood, of the amplitude or intensity of a SAR image.
The notion of local neighbouring is probably missing in the coef_var function. Based on this, I broke down the CoV function in two successive filtered band functions (Stddev and Mean) within a 5x5 window, applied to each of my amplitude images. I have then Averaged the results and visually got something consistent with what i was initially expecting from the Coef_var function. Maybe that’s a workaround.
A next step would be to “Graph” this with the Builder but it seems like I cannot apply from the “Image-Filter” block the same Non-Linear Filters from the “Filtered Band” tool such as a Standard Deviation 5x5 or a Mean 5x5. :confused:

So basically that is a textural measure it seems. BTW you might want to try including Intensity-Change between the two images as a feature in addition to the Coefficient of Variation, Mean amplitude and Coherence images you are already using.

@lveci should be able to help with the issue of calling Filtered Band - functionality from a Graph.

Yes, that would be worth trying.
And yes, having the Filtered Band function from the Graph would allow automation of a quite cumbersome process. Thank you.