While using ImageMagick patched with the original kurtosis patch, I realized I had made an error in magick/identify.c, where just like the mean and standard deviation, kurtosis and skewness were being divided by the values 'scale' and 'QuantumRange'.
While this makes sense for the mean and standard deviation, it does not in case of kurtosis and skewness. This is because the first two are dimensional numbers. They have units associated with them. So for the red channel the mean would be the mean amount of 'redness' as it were, etc. However kurtosis and skewness are dimensionless numbers, there are no units associated with them. See the following urls for an explanation: http://books.google.co.in/books?id=czbzO5iD1Z0C&pg=PA71&lpg=PA71&dq=kurtosis+dimensionless&source=web&ots=tR9hElz_yg&sig=SWORDeKlXaIl18Bsn2O-snEXdsc&hl=en&sa=X&oi=book_result&resnum=4&ct=result http://books.google.co.in/books?id=czbzO5iD1Z0C&pg=PA69&lpg=PA69&dq=skewness+dimensionless&source=web&ots=tR9hElATvf&sig=c09xpComTaj6zfJCV-HfTBY951w&hl=en&sa=X&oi=book_result&resnum=5&ct=result For this reason it makes no sense to divide them by 'scale' and 'QuantumRange'. I have made changes accordingly in magick/identify.c and also in PerlMagick/magick.xs, where too it makes no sense to divide kurtosis and skewness by 'scale'. I have also made a change in magick/identify.c to print the kurtosis and skewness for the image as a whole, i.e for AllChannels. It seems to be useful sometimes to know the overall values for these measures instead of just per channel. The changes in the other two files are spelling fixes, plus a reference to calculating kurtosis and skewness that I forget to put in the comments earlier. The url for the patch is: http://rapidshare.com/files/184401771/final_kurtosis_scaling_2.patch I hope this patch will be incorporated into ImageMagick. Please let me know what changes if any are required. Regards Amartyo _______________________________________________ Magick-developers mailing list Magick-developers@imagemagick.org http://studio.imagemagick.org/mailman/listinfo/magick-developers
