DBE and tolerance values

jeffrey merrill

Hi Adam:

your DBE use is a bit novel in your approach compared to others ...in that you seem to suggest (if I understood correctly) using whatever tolerance value is needed to achieve an adequate weighting for samples to get the full effect on the gradient....as long as your careful in placing samples and avoiding underlying stars etc.  Others have suggested much small increases in tolerance to just include samples known to be in background areas..therefore small increases in tolerance maybe using 2.0 - 2.5 or so..warning not to go too high

so I find your approach intriguing, especially since in the past I have struggled with removing some tough gradients..maybe I wasn't achieving the best model with lower tolerance values,  also maybe because I assumed most gradients in my flat calibrated images were additive in nature, especially in color images, but watching your tutorial I may have it wrong and it's multiplicative.

so if I understood correctly, and assuming I am very very careful with sample placement to really be in background areas, and avoid stars etc, is there any other downside to consider when using much higher tolerances (5,6, 8 etc)?

my other question is whether there's a downside to using more than one application of DBE on the same image, in terms of introducing noise or some other effect ...how far to you push it with one application vs 2 applications or more?

jeff

BlockHead

Yes, I think you are characterizing my philosophy accurately. No, I don't think there is a downside unless we are making quantitative measurements. The tolerance doesn't determine the degree/magnitude of the application of the model (in other words it isn't an aggressiveness parameter)- it does affect the calculation of the model of course- but with good samples... you will have an excellent result because the fit is good. 

Anecdotally Vicent Peris uses values of 10.

Yes, flat errors are definitely multiplicative errors (as this is how flats work). My argument is that most errors are flat-like errors and not additive light issues. 

Finally, I do not find it necessary to apply multiple applications of DBE in general. It is really for problem type images. One of the points I was trying to make is that complex gradients (whatever their proximate causes) are usually the combination of multiple simpler ones. So reduce the simplest ones until all you have a irreducible thing that you just throw samples at. 

More applications of DBE do not equate to better solutions... it isn't really a convergent process. It just the quality of model fit. 

-the Blockhead