Bob, I made that comment and appear to have made an unfortunate choice of words.
I meant "to minimize the contribution of the very strong pixel intensities due to large grains to the final intensity included in a diffraction pattern so that the final intensities more closely resemble those of a non-grainy sample" not the manner in which the particular reflection eventually contributes to the least squares procedure. (maybe I just compounded my sin?) Of course, both are eventually linked, so the manner in which applying a median filter to the 2-D image prior to producing a histogram affects the whole procedure is what I was wondering about. According to Jon's comments, there is a body of lore/practice associated with this procedure, but whether it has been discussed in print is uncertain. I must say that my experience in applying median filters of various styles (e.g., hybrid median filters) is pretty much limited to microscopy (AFM/SEM) image processing and never has been applied to XRD data. There certainly must be subtleties to such an implementation (e.g, the intensities are only expected to be similar in the direction of the diffraction ring, and the odd intensities are tens of pixels in dimension, etc.) Alex Y Dr. Alexandre (Alex) Yokochi Assistant Professor Department of Chemical Engineering Oregon State University Corvallis, OR 97331-2702 Phone: (541)-737-9357 Fax: (541)-737-4600 -----Original Message----- From: Von Dreele, Robert B. [mailto:[EMAIL PROTECTED] Sent: Tuesday, April 05, 2005 12:10 PM To: rietveld_l@ill.fr Re: Level of Preferred Orientation) Guys, This appeared in a Rietveld e-mail a bit ago & needs a comment: "While I can understand the general rationale for the idea (minimize the weight of the very strong reflections to the final integrated intensity for the reflection)" The fact of the matter is that most least squares programs doing Rietveld refinement (GSAS included) use weights that are equal to 1/I. This is what one expects from "pure" Poisson counting statistics for reasonably large numbers of counts (i.e. >20). Because of this choice of weights, each observation in the refinement is "equal" in terms of "impact" on the refinement. So the above suggestion that the weights for strong reflections be "reduced" is already done in the standard form of Rietveld least squares refinement. The "extra" residual you may see in the vicinity of strong reflections is actually no larger when weighted (unless the model isn't right, of course) than the surrounding lower intensity values. In GSAS, the graphics routine POWPLOT has an option that clearly illustrates this. The "W" option scales each intensity & difference by the weight. The resulting curve has no peaks(!) and is what the LS minimization engine actually sees for refinement. Bob Von Dreele R.B. Von Dreele IPNS Division Argonne National Laboratory Argonne, IL 60439-4814
