PJSR MAD functions?

[mschuster]

ImageIntegration's Scale + Zero Offset option uses a MAD function defined as 'median absolute deviation'.

Is this the 'median absolute deviation' from the 'mean' or from the 'median'?

Is there a PixelMath function implementing this? I can't find one in the documentation. If not, is there a way of implementing MAD as a more complex expression in pjsr somehow?

The Statistics process displays a 'AvgDev' value. I assume this is the 'mean absolute deviation' not 'median absolute deviation'. Is this correct? But again is it from the 'median' or the 'mean'?

Thanks,
Mike

[Geoff]

I would take it to mean the "median of the absolute deviation from the mean"
Geoff

[mschuster]

Thanks Geoff.

Here is my conclusion for Statistics AvgDev:

Running PixelMath Abs($target - median($target)) and then in Statistics look at the Mean of the result. This equals the AvgDev of the target.

So Statistics AvgDev is the 'mean absolute deviation from the median'.

So this is the reverse of your conclusion for MAD.

It is confusing that there are multiple variations of this type of function documented rather ambiguously.

Thanks,
Mike

[mschuster]

Oh no, my mistake, the documentation is clear. Clicking on the link 'median absolute deviation' in the ImageIntegration documentation for MAD brings up a Wiki page that defines MAD as 'median absolute deviation from the median'. I still don't see a PixelMath or ImageStatistics equivalent but I think I can code one up easily.
Regards,
Mike

[Juan Conejero]

AvgDev in Statistics, the avgdev function in PixelMath, and the Image.avgDev() and Math.avgDev() methods in PJSR, all of them refer to the mean absolute deviation from the median:

As for the MAD function (median absolute deviation from the median), there is no predefined method to compute it in PJSR. I'll implement it in version 1.8.0. It is indeed easy to write a JavaScript routine to calculate it.