


Improve performance of the pixel cache.If IPTC profile is not embedded in an 8bim resource, declare it IPTC rather than 8BIM.Removed LanczosFast from supported filters (SincFast*SincFast is faster).modified "magick/resample.c" to allow use of either EWA or HQEWA (default).Some computations were needlessly done in long double precision because PIL forced an automatic upcast wherever it appeared. Modified "magick/resize.c" so that MagickPIL is a MagickRealType number.Don't define HAVE_THREADS for -without-threads configure script option.Do not break words if caption size is absolute (reference ).Now that MagickPIL is a MagickRealType, some casts are unneeded.Don't allow resize filter weights to go to zero (reference ).Fix Point filter for ResizeImage() caused by support limiting the Box weighting function.This makes -distort a whole lot nicer and allows for the use of better cylindrical filters. Replace the blurry "High Quality EWA" technique with a 'Clamped EWA' for Distort Resampling.relative error version for Q64 as well as Q32. No reason to have "insane" precision Sinc in resize.c: Use the 1.2e-12 max.Simplified the ClampUpAxes code, the use of its outputs, its comments, and credited Craig DeForest for the "clamp singular values" idea.Bug fix for -filter point distorts (occasional generating black pixels).Temporary files not always deleted (reference ).Chantal Racette double checked the bounding parallelogram computation in resample.c (and I tightened the comments).'Bessel' filter is now offically and more accuritally named 'Jinc' however 'Bessel' while not visible as a filter option can still be used as an internal alias for 'Jinc'.Fix horizon anti-alising for output-scaled perspective distortions.Adjusted Variable Mapping Blur Composition so user arguments actual relate properly to the sigma of the blur for a maximum mapping value.Of course it is still a very blury filter for default use in EWA. This resulted removing the last aliasing issue that was present in tests for Gaussian EWA resampling. Added sqrt(2) bluring default for Gaussian Filter if used as a Cylindrical EWA filter.Don't permit access to pixels when pinging an image (reference ).Fix memory assertion with -enable-embeddable (reference ).Properly handle PSD layers with negative offsets.Add support for "pattern:vertical3" and "pattern:horizontal3".Add support for "pattern:vertical2" and "pattern:horizontal2".Added filter "Robidoux" which is a slightly sharpened version of the "Lanczos2D" filter (blur=0.958033808) specifically designed to be less 'blurry' for horizontal and vertical lines in no-op distortions.Added new filter 'Lanczos2D' a 2-lobe Lanczos as defined by Andreas Gustafsson in his thesis "Interactive Image Warping" (page 24).Autodetect file type support in Makefile.PL to prevent test failures.Original sharpened Lanczos2D filter now called "Lanczos2Dsharp".(C=0.31089212245300069) This also is very similer to a Mitchell filter but specifically designed for EWA use and is the new default filter for Distorting Images. Replaced "Robidoux" with Cubic 'Keys' filter that is near equivelent to the previous sharped "Lanczos2D" filter.The webp format requires the webpconv delegate program (experimental).

WHERE TO INSTALL ICC PROFILE MAC 10.12.6 PATCH
Patch for DrawableRotation() and DrawableTranslation() (reference ).(This will allow future adjustment of Gaussians for variable blur mapping) This is similar in action to 'blur' but only for Gaussians, and does not modify the filters support, allowing you to set a very small sigma, without the function 'missing' all pixels. Added "filter:sigma" expert setting defining the 'sigma' for the Gaussian filter only.Use #pragma omp master for dumping single threaded debugging information (single caused a fault).Ignore PS bounding box offsets if -page is set.Revert the Robidoux filter to a Keys cubic with C=(108 sqrt 2-29)/398 (as already specified in the documentation).In particular, P1 coefficient (always zero) removed from coeff. CubicBC filter formulas simplified by constant folding.Filter sharpening factors are also always applied regardless of usage.This is a rare case where a Jinc filter is also useful for resize. Lanczos2D and Lanczos2DSharp are now always a 2-lobe Jinc-Jinc filter regardless of orthoginal (resize) or cylindrical (distort-EWA) usage.Added LanczosSharp (3-lobe Lanczos with sharpening).Clearer EWA filters (LanczosSharp etc) comments.
