[EMAIL PROTECTED] on wrote... | We need to investigate further but ImageMagick appears to be slower | because it now uses a more sophisticated resampling algorithm-- for swirl, | implode, and wave-- rather than the previous interpolation algorithm. | | Anthony wrote the algorithm so he can explain why resampling is better | than straight interpolation. However, its possible to set the image | filter to revert to the previous interpolation algorithm. We will still | need a few days to investigate further. In the mean-time we did replace | the XML DOM parser with a SAX parser in 6.4.5-0 so that should reduce | some startup time. | | Will get back to you in a few days.
I never said it was better for those functions. In fact at the time the change occurred, I was against it being used so universally. But it was done before I know what had happened, it was working. It was also pointed out that the API interface was cleaner and more well defined, with proper initialization, working functions, and finalization methods. The older method by contrast used methods from multiple areas of the IM API to initialize and finalize the proceedures. The Details... The Resampling algorithm was a wrapper around the old "Interpolated Pixel Lookup" function (single value lookup from the source image), I created it for use by the new image distortion functions which was being co-developed at the same time. See the whole section in IM Examples: General Distortion Techniques http://www.imagemagick.org/Usage/distorts/#summary for an explaination o fthe various techniques involved. Basically it takes not only the point of the lookup, but also 4 scaling factors (actually two scaling vectors) which it uses to do a cylindrically filters elliptical area, when the scaling is larger than one. That is it allows for correct anti-aliasing filtering in areas where a distortion shrinks an image into a smaller area, especially in the extreme case of tiled perspective distortion. Special Thanks goes to Fred Weinhaus for his help with the algorithm. This is in actal fact very simular to Imge Resizing (Shrinking Images) except that resize uses a two pass one dimensional filter, where area resampling is a single pass 2 dimentional cylindrical filter, with variable (localized) scaling. This is also slower than resize but more freeform whcih is needed for generalized distortions. It isn't even quite finished as it is currently using a blurry gaussian filter which was part of the original algorithms definition, by default, and while other filters can be used, the filters have a 'limited' suport factor handling (to get technical). This multiple point area lookup is automatically disabled if the scaling factors are not used (only -distort uses them), the scaling is smaller than 1 (enlargement), or the setting -filter point is set. In this case it just calls the old interpolated pixel lookup. However the algorthim still needs fee form access to the image requiring some extra setup and handling of the image chache. Which is where the slower speed is comming in, even if the area resampling aspect is NOT used. After I added the Area resampling function, and with some mis-givings by myself, most of the other functions were modified to start using this 'wrapper', even though none of those functions even needed area resampling, only the point lookup. At this time ONLY DistortImage() actually needs to perform area-resampling. Though eventually I would expect other functions like the planned replacement for Distortion and Displacement LUT Mapping to also use this method. All other operators, which are only doing single 'point' lookups, does not need to use Area Resampling, and could be changed back to the direct use of the simpler Interpolated Point Sampling. This will improve IM speed back to previous levels. Anthony Thyssen ( System Programmer ) <[EMAIL PROTECTED]> ----------------------------------------------------------------------------- "Never argue with a computer.' -- Avon, "Blake's 7" ----------------------------------------------------------------------------- Anthony's Home is his Castle http://www.cit.gu.edu.au/~anthony/ _______________________________________________ Magick-developers mailing list Magick-developers@imagemagick.org http://studio.imagemagick.org/mailman/listinfo/magick-developers
