2 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
6 % RRRR EEEEE SSSSS AAA M M PPPP L EEEEE %
7 % R R E SS A A MM MM P P L E %
8 % RRRR EEE SSS AAAAA M M M PPPP L EEE %
9 % R R E SS A A M M P L E %
10 % R R EEEEE SSSSS A A M M P LLLLL EEEEE %
13 % MagickCore Pixel Resampling Methods %
21 % Copyright 1999-2010 ImageMagick Studio LLC, a non-profit organization %
22 % dedicated to making software imaging solutions freely available. %
24 % You may not use this file except in compliance with the License. You may %
25 % obtain a copy of the License at %
27 % http://www.imagemagick.org/script/license.php %
29 % Unless required by applicable law or agreed to in writing, software %
30 % distributed under the License is distributed on an "AS IS" BASIS, %
31 % WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. %
32 % See the License for the specific language governing permissions and %
33 % limitations under the License. %
35 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
43 #include "magick/studio.h"
44 #include "magick/artifact.h"
45 #include "magick/color-private.h"
46 #include "magick/cache.h"
47 #include "magick/draw.h"
48 #include "magick/exception-private.h"
49 #include "magick/gem.h"
50 #include "magick/image.h"
51 #include "magick/image-private.h"
52 #include "magick/log.h"
53 #include "magick/magick.h"
54 #include "magick/memory_.h"
55 #include "magick/pixel-private.h"
56 #include "magick/quantum.h"
57 #include "magick/random_.h"
58 #include "magick/resample.h"
59 #include "magick/resize.h"
60 #include "magick/resize-private.h"
61 #include "magick/transform.h"
62 #include "magick/signature-private.h"
63 #include "magick/utility.h"
65 EWA Resampling Options
68 /* select ONE resampling method */
69 #define EWA 1 /* Normal EWA handling - raw or clamped */
70 /* if 0 then use "High Quality EWA" */
71 #define EWA_CLAMP 1 /* EWA Clamping from Nicolas Robidoux */
73 #define FILTER_LUT 1 /* Use a LUT rather then direct filter calls */
75 /* output debugging information */
76 #define DEBUG_ELLIPSE 0 /* output ellipse info for debug */
77 #define DEBUG_HIT_MISS 0 /* output hit/miss pixels (as gnuplot commands) */
78 #define DEBUG_NO_PIXEL_HIT 0 /* Make pixels that fail to hit anything - RED */
81 #define WLUT_WIDTH 1024 /* size of the filter cache */
87 struct _ResampleFilter
101 /* Information about image being resampled */
105 InterpolatePixelMethod
114 /* processing settings needed */
123 /* current ellipitical area being resampled around center point */
126 Vlimit, Ulimit, Uwidth, slope;
129 /* LUT of weights for filtered average in elliptical area */
131 filter_lut[WLUT_WIDTH];
133 /* Use a Direct call to the filter functions */
141 /* the practical working support of the filter */
150 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
154 % A c q u i r e R e s a m p l e I n f o %
158 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
160 % AcquireResampleFilter() initializes the information resample needs do to a
161 % scaled lookup of a color from an image, using area sampling.
163 % The algorithm is based on a Elliptical Weighted Average, where the pixels
164 % found in a large elliptical area is averaged together according to a
165 % weighting (filter) function. For more details see "Fundamentals of Texture
166 % Mapping and Image Warping" a master's thesis by Paul.S.Heckbert, June 17,
167 % 1989. Available for free from, http://www.cs.cmu.edu/~ph/
169 % As EWA resampling (or any sort of resampling) can require a lot of
170 % calculations to produce a distorted scaling of the source image for each
171 % output pixel, the ResampleFilter structure generated holds that information
172 % between individual image resampling.
174 % This function will make the appropriate AcquireCacheView() calls
175 % to view the image, calling functions do not need to open a cache view.
178 % resample_filter=AcquireResampleFilter(image,exception);
179 % SetResampleFilter(resample_filter, GaussianFilter, 1.0);
180 % for (y=0; y < (ssize_t) image->rows; y++) {
181 % for (x=0; x < (ssize_t) image->columns; x++) {
183 % ScaleResampleFilter(resample_filter, ... scaling vectors ...);
184 % (void) ResamplePixelColor(resample_filter,u,v,&pixel);
185 % ... assign resampled pixel value ...
188 % DestroyResampleFilter(resample_filter);
190 % The format of the AcquireResampleFilter method is:
192 % ResampleFilter *AcquireResampleFilter(const Image *image,
193 % ExceptionInfo *exception)
195 % A description of each parameter follows:
197 % o image: the image.
199 % o exception: return any errors or warnings in this structure.
202 MagickExport ResampleFilter *AcquireResampleFilter(const Image *image,
203 ExceptionInfo *exception)
205 register ResampleFilter
208 assert(image != (Image *) NULL);
209 assert(image->signature == MagickSignature);
210 if (image->debug != MagickFalse)
211 (void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
212 assert(exception != (ExceptionInfo *) NULL);
213 assert(exception->signature == MagickSignature);
215 resample_filter=(ResampleFilter *) AcquireMagickMemory(
216 sizeof(*resample_filter));
217 if (resample_filter == (ResampleFilter *) NULL)
218 ThrowFatalException(ResourceLimitFatalError,"MemoryAllocationFailed");
219 (void) ResetMagickMemory(resample_filter,0,sizeof(*resample_filter));
221 resample_filter->image=ReferenceImage((Image *) image);
222 resample_filter->view=AcquireCacheView(resample_filter->image);
223 resample_filter->exception=exception;
225 resample_filter->debug=IsEventLogging();
226 resample_filter->signature=MagickSignature;
228 resample_filter->image_area=(ssize_t) (image->columns*image->rows);
229 resample_filter->average_defined = MagickFalse;
231 /* initialise the resampling filter settings */
232 SetResampleFilter(resample_filter, image->filter, image->blur);
233 (void) SetResampleFilterInterpolateMethod(resample_filter,
235 (void) SetResampleFilterVirtualPixelMethod(resample_filter,
236 GetImageVirtualPixelMethod(image));
238 return(resample_filter);
242 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
246 % D e s t r o y R e s a m p l e I n f o %
250 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
252 % DestroyResampleFilter() finalizes and cleans up the resampling
253 % resample_filter as returned by AcquireResampleFilter(), freeing any memory
254 % or other information as needed.
256 % The format of the DestroyResampleFilter method is:
258 % ResampleFilter *DestroyResampleFilter(ResampleFilter *resample_filter)
260 % A description of each parameter follows:
262 % o resample_filter: resampling information structure
265 MagickExport ResampleFilter *DestroyResampleFilter(
266 ResampleFilter *resample_filter)
268 assert(resample_filter != (ResampleFilter *) NULL);
269 assert(resample_filter->signature == MagickSignature);
270 assert(resample_filter->image != (Image *) NULL);
271 if (resample_filter->debug != MagickFalse)
272 (void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",
273 resample_filter->image->filename);
274 resample_filter->view=DestroyCacheView(resample_filter->view);
275 resample_filter->image=DestroyImage(resample_filter->image);
277 resample_filter->filter_def=DestroyResizeFilter(resample_filter->filter_def);
279 resample_filter->signature=(~MagickSignature);
280 resample_filter=(ResampleFilter *) RelinquishMagickMemory(resample_filter);
281 return(resample_filter);
285 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
289 % I n t e r p o l a t e R e s a m p l e F i l t e r %
293 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
295 % InterpolateResampleFilter() applies bi-linear or tri-linear interpolation
296 % between a floating point coordinate and the pixels surrounding that
297 % coordinate. No pixel area resampling, or scaling of the result is
300 % The format of the InterpolateResampleFilter method is:
302 % MagickBooleanType InterpolateResampleFilter(
303 % ResampleInfo *resample_filter,const InterpolatePixelMethod method,
304 % const double x,const double y,MagickPixelPacket *pixel)
306 % A description of each parameter follows:
308 % o resample_filter: the resample filter.
310 % o method: the pixel clor interpolation method.
312 % o x,y: A double representing the current (x,y) position of the pixel.
314 % o pixel: return the interpolated pixel here.
318 static inline double MagickMax(const double x,const double y)
325 static void BicubicInterpolate(const MagickPixelPacket *pixels,const double dx,
326 MagickPixelPacket *pixel)
336 p=(pixels[3].red-pixels[2].red)-(pixels[0].red-pixels[1].red);
337 q=(pixels[0].red-pixels[1].red)-p;
338 r=pixels[2].red-pixels[0].red;
340 pixel->red=(dx*dx2*p)+(dx2*q)+(dx*r)+s;
341 p=(pixels[3].green-pixels[2].green)-(pixels[0].green-pixels[1].green);
342 q=(pixels[0].green-pixels[1].green)-p;
343 r=pixels[2].green-pixels[0].green;
345 pixel->green=(dx*dx2*p)+(dx2*q)+(dx*r)+s;
346 p=(pixels[3].blue-pixels[2].blue)-(pixels[0].blue-pixels[1].blue);
347 q=(pixels[0].blue-pixels[1].blue)-p;
348 r=pixels[2].blue-pixels[0].blue;
350 pixel->blue=(dx*dx2*p)+(dx2*q)+(dx*r)+s;
351 p=(pixels[3].opacity-pixels[2].opacity)-(pixels[0].opacity-pixels[1].opacity);
352 q=(pixels[0].opacity-pixels[1].opacity)-p;
353 r=pixels[2].opacity-pixels[0].opacity;
355 pixel->opacity=(dx*dx2*p)+(dx2*q)+(dx*r)+s;
356 if (pixel->colorspace == CMYKColorspace)
358 p=(pixels[3].index-pixels[2].index)-(pixels[0].index-pixels[1].index);
359 q=(pixels[0].index-pixels[1].index)-p;
360 r=pixels[2].index-pixels[0].index;
362 pixel->index=(dx*dx2*p)+(dx2*q)+(dx*r)+s;
366 static inline MagickRealType CubicWeightingFunction(const MagickRealType x)
372 alpha=MagickMax(x+2.0,0.0);
373 gamma=1.0*alpha*alpha*alpha;
374 alpha=MagickMax(x+1.0,0.0);
375 gamma-=4.0*alpha*alpha*alpha;
376 alpha=MagickMax(x+0.0,0.0);
377 gamma+=6.0*alpha*alpha*alpha;
378 alpha=MagickMax(x-1.0,0.0);
379 gamma-=4.0*alpha*alpha*alpha;
383 static inline double MeshInterpolate(const PointInfo *delta,const double p,
384 const double x,const double y)
386 return(delta->x*x+delta->y*y+(1.0-delta->x-delta->y)*p);
389 static inline ssize_t NearestNeighbor(MagickRealType x)
392 return((ssize_t) (x+0.5));
393 return((ssize_t) (x-0.5));
396 static MagickBooleanType InterpolateResampleFilter(
397 ResampleFilter *resample_filter,const InterpolatePixelMethod method,
398 const double x,const double y,MagickPixelPacket *pixel)
403 register const IndexPacket
406 register const PixelPacket
412 assert(resample_filter != (ResampleFilter *) NULL);
413 assert(resample_filter->signature == MagickSignature);
418 case AverageInterpolatePixel:
427 p=GetCacheViewVirtualPixels(resample_filter->view,(ssize_t) floor(x)-1,
428 (ssize_t) floor(y)-1,4,4,resample_filter->exception);
429 if (p == (const PixelPacket *) NULL)
434 indexes=GetCacheViewVirtualIndexQueue(resample_filter->view);
435 for (i=0; i < 16L; i++)
437 GetMagickPixelPacket(resample_filter->image,pixels+i);
438 SetMagickPixelPacket(resample_filter->image,p,indexes+i,pixels+i);
440 if (resample_filter->image->matte != MagickFalse)
442 alpha[i]=QuantumScale*((MagickRealType) GetAlphaPixelComponent(p));
443 pixels[i].red*=alpha[i];
444 pixels[i].green*=alpha[i];
445 pixels[i].blue*=alpha[i];
446 if (resample_filter->image->colorspace == CMYKColorspace)
447 pixels[i].index*=alpha[i];
450 gamma=1.0/(fabs((double) gamma) <= MagickEpsilon ? 1.0 : gamma);
451 pixel->red+=gamma*0.0625*pixels[i].red;
452 pixel->green+=gamma*0.0625*pixels[i].green;
453 pixel->blue+=gamma*0.0625*pixels[i].blue;
454 pixel->opacity+=0.0625*pixels[i].opacity;
455 if (resample_filter->image->colorspace == CMYKColorspace)
456 pixel->index+=gamma*0.0625*pixels[i].index;
461 case BicubicInterpolatePixel:
473 p=GetCacheViewVirtualPixels(resample_filter->view,(ssize_t) floor(x)-1,
474 (ssize_t) floor(y)-1,4,4,resample_filter->exception);
475 if (p == (const PixelPacket *) NULL)
480 indexes=GetCacheViewVirtualIndexQueue(resample_filter->view);
481 for (i=0; i < 16L; i++)
483 GetMagickPixelPacket(resample_filter->image,pixels+i);
484 SetMagickPixelPacket(resample_filter->image,p,indexes+i,pixels+i);
486 if (resample_filter->image->matte != MagickFalse)
488 alpha[i]=QuantumScale*((MagickRealType) GetAlphaPixelComponent(p));
489 pixels[i].red*=alpha[i];
490 pixels[i].green*=alpha[i];
491 pixels[i].blue*=alpha[i];
492 if (resample_filter->image->colorspace == CMYKColorspace)
493 pixels[i].index*=alpha[i];
498 for (i=0; i < 4L; i++)
499 BicubicInterpolate(pixels+4*i,delta.x,u+i);
501 BicubicInterpolate(u,delta.y,pixel);
504 case BilinearInterpolatePixel:
518 p=GetCacheViewVirtualPixels(resample_filter->view,(ssize_t) floor(x),
519 (ssize_t) floor(y),2,2,resample_filter->exception);
520 if (p == (const PixelPacket *) NULL)
525 indexes=GetCacheViewVirtualIndexQueue(resample_filter->view);
526 for (i=0; i < 4L; i++)
528 pixels[i].red=(MagickRealType) p[i].red;
529 pixels[i].green=(MagickRealType) p[i].green;
530 pixels[i].blue=(MagickRealType) p[i].blue;
531 pixels[i].opacity=(MagickRealType) p[i].opacity;
534 if (resample_filter->image->matte != MagickFalse)
535 for (i=0; i < 4L; i++)
537 alpha[i]=QuantumScale*((MagickRealType) QuantumRange-p[i].opacity);
538 pixels[i].red*=alpha[i];
539 pixels[i].green*=alpha[i];
540 pixels[i].blue*=alpha[i];
542 if (indexes != (IndexPacket *) NULL)
543 for (i=0; i < 4L; i++)
545 pixels[i].index=(MagickRealType) indexes[i];
546 if (resample_filter->image->colorspace == CMYKColorspace)
547 pixels[i].index*=alpha[i];
551 epsilon.x=1.0-delta.x;
552 epsilon.y=1.0-delta.y;
553 gamma=((epsilon.y*(epsilon.x*alpha[0]+delta.x*alpha[1])+delta.y*
554 (epsilon.x*alpha[2]+delta.x*alpha[3])));
555 gamma=1.0/(fabs((double) gamma) <= MagickEpsilon ? 1.0 : gamma);
556 pixel->red=gamma*(epsilon.y*(epsilon.x*pixels[0].red+delta.x*
557 pixels[1].red)+delta.y*(epsilon.x*pixels[2].red+delta.x*pixels[3].red));
558 pixel->green=gamma*(epsilon.y*(epsilon.x*pixels[0].green+delta.x*
559 pixels[1].green)+delta.y*(epsilon.x*pixels[2].green+delta.x*
561 pixel->blue=gamma*(epsilon.y*(epsilon.x*pixels[0].blue+delta.x*
562 pixels[1].blue)+delta.y*(epsilon.x*pixels[2].blue+delta.x*
564 pixel->opacity=(epsilon.y*(epsilon.x*pixels[0].opacity+delta.x*
565 pixels[1].opacity)+delta.y*(epsilon.x*pixels[2].opacity+delta.x*
567 if (resample_filter->image->colorspace == CMYKColorspace)
568 pixel->index=gamma*(epsilon.y*(epsilon.x*pixels[0].index+delta.x*
569 pixels[1].index)+delta.y*(epsilon.x*pixels[2].index+delta.x*
573 case FilterInterpolatePixel:
590 geometry.x=(ssize_t) floor(x)-1L;
591 geometry.y=(ssize_t) floor(y)-1L;
592 excerpt_image=ExcerptImage(resample_filter->image,&geometry,
593 resample_filter->exception);
594 if (excerpt_image == (Image *) NULL)
599 filter_image=ResizeImage(excerpt_image,1,1,resample_filter->image->filter,
600 resample_filter->image->blur,resample_filter->exception);
601 excerpt_image=DestroyImage(excerpt_image);
602 if (filter_image == (Image *) NULL)
604 filter_view=AcquireCacheView(filter_image);
605 p=GetCacheViewVirtualPixels(filter_view,0,0,1,1,
606 resample_filter->exception);
607 if (p != (const PixelPacket *) NULL)
609 indexes=GetVirtualIndexQueue(filter_image);
610 GetMagickPixelPacket(resample_filter->image,pixels);
611 SetMagickPixelPacket(resample_filter->image,p,indexes,pixel);
613 filter_view=DestroyCacheView(filter_view);
614 filter_image=DestroyImage(filter_image);
617 case IntegerInterpolatePixel:
622 p=GetCacheViewVirtualPixels(resample_filter->view,(ssize_t) floor(x),
623 (ssize_t) floor(y),1,1,resample_filter->exception);
624 if (p == (const PixelPacket *) NULL)
629 indexes=GetCacheViewVirtualIndexQueue(resample_filter->view);
630 GetMagickPixelPacket(resample_filter->image,pixels);
631 SetMagickPixelPacket(resample_filter->image,p,indexes,pixel);
634 case MeshInterpolatePixel:
647 p=GetCacheViewVirtualPixels(resample_filter->view,(ssize_t) floor(x),
648 (ssize_t) floor(y),2,2,resample_filter->exception);
649 if (p == (const PixelPacket *) NULL)
654 indexes=GetCacheViewVirtualIndexQueue(resample_filter->view);
655 for (i=0; i < 4L; i++)
657 GetMagickPixelPacket(resample_filter->image,pixels+i);
658 SetMagickPixelPacket(resample_filter->image,p,indexes+i,pixels+i);
660 if (resample_filter->image->matte != MagickFalse)
662 alpha[i]=QuantumScale*((MagickRealType) GetAlphaPixelComponent(p));
663 pixels[i].red*=alpha[i];
664 pixels[i].green*=alpha[i];
665 pixels[i].blue*=alpha[i];
666 if (resample_filter->image->colorspace == CMYKColorspace)
667 pixels[i].index*=alpha[i];
673 luminance.x=MagickPixelLuminance(pixels+0)-MagickPixelLuminance(pixels+3);
674 luminance.y=MagickPixelLuminance(pixels+1)-MagickPixelLuminance(pixels+2);
675 if (fabs(luminance.x) < fabs(luminance.y))
680 if (delta.x <= delta.y)
683 Bottom-left triangle (pixel:2, diagonal: 0-3).
686 gamma=MeshInterpolate(&delta,alpha[2],alpha[3],alpha[0]);
687 gamma=1.0/(fabs((double) gamma) <= MagickEpsilon ? 1.0 : gamma);
688 pixel->red=gamma*MeshInterpolate(&delta,pixels[2].red,
689 pixels[3].red,pixels[0].red);
690 pixel->green=gamma*MeshInterpolate(&delta,pixels[2].green,
691 pixels[3].green,pixels[0].green);
692 pixel->blue=gamma*MeshInterpolate(&delta,pixels[2].blue,
693 pixels[3].blue,pixels[0].blue);
694 pixel->opacity=gamma*MeshInterpolate(&delta,pixels[2].opacity,
695 pixels[3].opacity,pixels[0].opacity);
696 if (resample_filter->image->colorspace == CMYKColorspace)
697 pixel->index=gamma*MeshInterpolate(&delta,pixels[2].index,
698 pixels[3].index,pixels[0].index);
703 Top-right triangle (pixel:1, diagonal: 0-3).
706 gamma=MeshInterpolate(&delta,alpha[1],alpha[0],alpha[3]);
707 gamma=1.0/(fabs((double) gamma) <= MagickEpsilon ? 1.0 : gamma);
708 pixel->red=gamma*MeshInterpolate(&delta,pixels[1].red,
709 pixels[0].red,pixels[3].red);
710 pixel->green=gamma*MeshInterpolate(&delta,pixels[1].green,
711 pixels[0].green,pixels[3].green);
712 pixel->blue=gamma*MeshInterpolate(&delta,pixels[1].blue,
713 pixels[0].blue,pixels[3].blue);
714 pixel->opacity=gamma*MeshInterpolate(&delta,pixels[1].opacity,
715 pixels[0].opacity,pixels[3].opacity);
716 if (resample_filter->image->colorspace == CMYKColorspace)
717 pixel->index=gamma*MeshInterpolate(&delta,pixels[1].index,
718 pixels[0].index,pixels[3].index);
726 if (delta.x <= (1.0-delta.y))
729 Top-left triangle (pixel 0, diagonal: 1-2).
731 gamma=MeshInterpolate(&delta,alpha[0],alpha[1],alpha[2]);
732 gamma=1.0/(fabs((double) gamma) <= MagickEpsilon ? 1.0 : gamma);
733 pixel->red=gamma*MeshInterpolate(&delta,pixels[0].red,
734 pixels[1].red,pixels[2].red);
735 pixel->green=gamma*MeshInterpolate(&delta,pixels[0].green,
736 pixels[1].green,pixels[2].green);
737 pixel->blue=gamma*MeshInterpolate(&delta,pixels[0].blue,
738 pixels[1].blue,pixels[2].blue);
739 pixel->opacity=gamma*MeshInterpolate(&delta,pixels[0].opacity,
740 pixels[1].opacity,pixels[2].opacity);
741 if (resample_filter->image->colorspace == CMYKColorspace)
742 pixel->index=gamma*MeshInterpolate(&delta,pixels[0].index,
743 pixels[1].index,pixels[2].index);
748 Bottom-right triangle (pixel: 3, diagonal: 1-2).
752 gamma=MeshInterpolate(&delta,alpha[3],alpha[2],alpha[1]);
753 gamma=1.0/(fabs((double) gamma) <= MagickEpsilon ? 1.0 : gamma);
754 pixel->red=gamma*MeshInterpolate(&delta,pixels[3].red,
755 pixels[2].red,pixels[1].red);
756 pixel->green=gamma*MeshInterpolate(&delta,pixels[3].green,
757 pixels[2].green,pixels[1].green);
758 pixel->blue=gamma*MeshInterpolate(&delta,pixels[3].blue,
759 pixels[2].blue,pixels[1].blue);
760 pixel->opacity=gamma*MeshInterpolate(&delta,pixels[3].opacity,
761 pixels[2].opacity,pixels[1].opacity);
762 if (resample_filter->image->colorspace == CMYKColorspace)
763 pixel->index=gamma*MeshInterpolate(&delta,pixels[3].index,
764 pixels[2].index,pixels[1].index);
769 case NearestNeighborInterpolatePixel:
774 p=GetCacheViewVirtualPixels(resample_filter->view,NearestNeighbor(x),
775 NearestNeighbor(y),1,1,resample_filter->exception);
776 if (p == (const PixelPacket *) NULL)
781 indexes=GetCacheViewVirtualIndexQueue(resample_filter->view);
782 GetMagickPixelPacket(resample_filter->image,pixels);
783 SetMagickPixelPacket(resample_filter->image,p,indexes,pixel);
786 case SplineInterpolatePixel:
804 p=GetCacheViewVirtualPixels(resample_filter->view,(ssize_t) floor(x)-1,
805 (ssize_t) floor(y)-1,4,4,resample_filter->exception);
806 if (p == (const PixelPacket *) NULL)
811 indexes=GetCacheViewVirtualIndexQueue(resample_filter->view);
815 for (i=(-1); i < 3L; i++)
817 dy=CubicWeightingFunction((MagickRealType) i-delta.y);
818 for (j=(-1); j < 3L; j++)
820 GetMagickPixelPacket(resample_filter->image,pixels+n);
821 SetMagickPixelPacket(resample_filter->image,p,indexes+n,pixels+n);
823 if (resample_filter->image->matte != MagickFalse)
825 alpha[n]=QuantumScale*((MagickRealType)
826 GetAlphaPixelComponent(p));
827 pixels[n].red*=alpha[n];
828 pixels[n].green*=alpha[n];
829 pixels[n].blue*=alpha[n];
830 if (resample_filter->image->colorspace == CMYKColorspace)
831 pixels[n].index*=alpha[n];
833 dx=CubicWeightingFunction(delta.x-(MagickRealType) j);
835 gamma=1.0/(fabs((double) gamma) <= MagickEpsilon ? 1.0 : gamma);
836 pixel->red+=gamma*dx*dy*pixels[n].red;
837 pixel->green+=gamma*dx*dy*pixels[n].green;
838 pixel->blue+=gamma*dx*dy*pixels[n].blue;
839 if (resample_filter->image->matte != MagickFalse)
840 pixel->opacity+=dx*dy*pixels[n].opacity;
841 if (resample_filter->image->colorspace == CMYKColorspace)
842 pixel->index+=gamma*dx*dy*pixels[n].index;
854 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
858 % R e s a m p l e P i x e l C o l o r %
862 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
864 % ResamplePixelColor() samples the pixel values surrounding the location
865 % given using an elliptical weighted average, at the scale previously
866 % calculated, and in the most efficent manner possible for the
867 % VirtualPixelMethod setting.
869 % The format of the ResamplePixelColor method is:
871 % MagickBooleanType ResamplePixelColor(ResampleFilter *resample_filter,
872 % const double u0,const double v0,MagickPixelPacket *pixel)
874 % A description of each parameter follows:
876 % o resample_filter: the resample filter.
878 % o u0,v0: A double representing the center of the area to resample,
879 % The distortion transformed transformed x,y coordinate.
881 % o pixel: the resampled pixel is returned here.
884 MagickExport MagickBooleanType ResamplePixelColor(
885 ResampleFilter *resample_filter,const double u0,const double v0,
886 MagickPixelPacket *pixel)
891 ssize_t u,v, v1, v2, uw, hit;
894 double divisor_c,divisor_m;
895 register double weight;
896 register const PixelPacket *pixels;
897 register const IndexPacket *indexes;
898 assert(resample_filter != (ResampleFilter *) NULL);
899 assert(resample_filter->signature == MagickSignature);
902 GetMagickPixelPacket(resample_filter->image,pixel);
903 if ( resample_filter->do_interpolate ) {
904 status=InterpolateResampleFilter(resample_filter,
905 resample_filter->interpolate,u0,v0,pixel);
910 fprintf(stderr, "u0=%lf; v0=%lf;\n", u0, v0);
914 Does resample area Miss the image?
915 And is that area a simple solid color - then return that color
918 switch ( resample_filter->virtual_pixel ) {
919 case BackgroundVirtualPixelMethod:
920 case ConstantVirtualPixelMethod:
921 case TransparentVirtualPixelMethod:
922 case BlackVirtualPixelMethod:
923 case GrayVirtualPixelMethod:
924 case WhiteVirtualPixelMethod:
925 case MaskVirtualPixelMethod:
926 if ( resample_filter->limit_reached
927 || u0 + resample_filter->Ulimit < 0.0
928 || u0 - resample_filter->Ulimit > (double) resample_filter->image->columns
929 || v0 + resample_filter->Vlimit < 0.0
930 || v0 - resample_filter->Vlimit > (double) resample_filter->image->rows
935 case UndefinedVirtualPixelMethod:
936 case EdgeVirtualPixelMethod:
937 if ( ( u0 + resample_filter->Ulimit < 0.0 && v0 + resample_filter->Vlimit < 0.0 )
938 || ( u0 + resample_filter->Ulimit < 0.0
939 && v0 - resample_filter->Vlimit > (double) resample_filter->image->rows )
940 || ( u0 - resample_filter->Ulimit > (double) resample_filter->image->columns
941 && v0 + resample_filter->Vlimit < 0.0 )
942 || ( u0 - resample_filter->Ulimit > (double) resample_filter->image->columns
943 && v0 - resample_filter->Vlimit > (double) resample_filter->image->rows )
947 case HorizontalTileVirtualPixelMethod:
948 if ( v0 + resample_filter->Vlimit < 0.0
949 || v0 - resample_filter->Vlimit > (double) resample_filter->image->rows
951 hit++; /* outside the horizontally tiled images. */
953 case VerticalTileVirtualPixelMethod:
954 if ( u0 + resample_filter->Ulimit < 0.0
955 || u0 - resample_filter->Ulimit > (double) resample_filter->image->columns
957 hit++; /* outside the vertically tiled images. */
959 case DitherVirtualPixelMethod:
960 if ( ( u0 + resample_filter->Ulimit < -32.0 && v0 + resample_filter->Vlimit < -32.0 )
961 || ( u0 + resample_filter->Ulimit < -32.0
962 && v0 - resample_filter->Vlimit > (double) resample_filter->image->rows+32.0 )
963 || ( u0 - resample_filter->Ulimit > (double) resample_filter->image->columns+32.0
964 && v0 + resample_filter->Vlimit < -32.0 )
965 || ( u0 - resample_filter->Ulimit > (double) resample_filter->image->columns+32.0
966 && v0 - resample_filter->Vlimit > (double) resample_filter->image->rows+32.0 )
970 case TileVirtualPixelMethod:
971 case MirrorVirtualPixelMethod:
972 case RandomVirtualPixelMethod:
973 case HorizontalTileEdgeVirtualPixelMethod:
974 case VerticalTileEdgeVirtualPixelMethod:
975 case CheckerTileVirtualPixelMethod:
976 /* resampling of area is always needed - no VP limits */
980 /* whole area is a solid color -- just return that color */
981 status=InterpolateResampleFilter(resample_filter,IntegerInterpolatePixel,
987 Scaling limits reached, return an 'averaged' result.
989 if ( resample_filter->limit_reached ) {
990 switch ( resample_filter->virtual_pixel ) {
991 /* This is always handled by the above, so no need.
992 case BackgroundVirtualPixelMethod:
993 case ConstantVirtualPixelMethod:
994 case TransparentVirtualPixelMethod:
995 case GrayVirtualPixelMethod,
996 case WhiteVirtualPixelMethod
997 case MaskVirtualPixelMethod:
999 case UndefinedVirtualPixelMethod:
1000 case EdgeVirtualPixelMethod:
1001 case DitherVirtualPixelMethod:
1002 case HorizontalTileEdgeVirtualPixelMethod:
1003 case VerticalTileEdgeVirtualPixelMethod:
1004 /* We need an average edge pixel, from the correct edge!
1005 How should I calculate an average edge color?
1006 Just returning an averaged neighbourhood,
1007 works well in general, but falls down for TileEdge methods.
1008 This needs to be done properly!!!!!!
1010 status=InterpolateResampleFilter(resample_filter,
1011 AverageInterpolatePixel,u0,v0,pixel);
1013 case HorizontalTileVirtualPixelMethod:
1014 case VerticalTileVirtualPixelMethod:
1015 /* just return the background pixel - Is there more direct way? */
1016 status=InterpolateResampleFilter(resample_filter,
1017 IntegerInterpolatePixel,(double)-1,(double)-1,pixel);
1019 case TileVirtualPixelMethod:
1020 case MirrorVirtualPixelMethod:
1021 case RandomVirtualPixelMethod:
1022 case CheckerTileVirtualPixelMethod:
1024 /* generate a average color of the WHOLE image */
1025 if ( resample_filter->average_defined == MagickFalse ) {
1032 GetMagickPixelPacket(resample_filter->image,(MagickPixelPacket *)
1033 &resample_filter->average_pixel);
1034 resample_filter->average_defined=MagickTrue;
1036 /* Try to get an averaged pixel color of whole image */
1037 average_image=ResizeImage(resample_filter->image,1,1,BoxFilter,1.0,
1038 resample_filter->exception);
1039 if (average_image == (Image *) NULL)
1041 *pixel=resample_filter->average_pixel; /* FAILED */
1044 average_view=AcquireCacheView(average_image);
1045 pixels=(PixelPacket *)GetCacheViewVirtualPixels(average_view,0,0,1,1,
1046 resample_filter->exception);
1047 if (pixels == (const PixelPacket *) NULL) {
1048 average_view=DestroyCacheView(average_view);
1049 average_image=DestroyImage(average_image);
1050 *pixel=resample_filter->average_pixel; /* FAILED */
1053 indexes=(IndexPacket *) GetCacheViewAuthenticIndexQueue(average_view);
1054 SetMagickPixelPacket(resample_filter->image,pixels,indexes,
1055 &(resample_filter->average_pixel));
1056 average_view=DestroyCacheView(average_view);
1057 average_image=DestroyImage(average_image);
1059 if ( resample_filter->virtual_pixel == CheckerTileVirtualPixelMethod )
1061 /* CheckerTile is avergae of image average half background */
1062 /* FUTURE: replace with a 50% blend of both pixels */
1064 weight = QuantumScale*((MagickRealType)(QuantumRange-
1065 resample_filter->average_pixel.opacity));
1066 resample_filter->average_pixel.red *= weight;
1067 resample_filter->average_pixel.green *= weight;
1068 resample_filter->average_pixel.blue *= weight;
1071 weight = QuantumScale*((MagickRealType)(QuantumRange-
1072 resample_filter->image->background_color.opacity));
1073 resample_filter->average_pixel.red +=
1074 weight*resample_filter->image->background_color.red;
1075 resample_filter->average_pixel.green +=
1076 weight*resample_filter->image->background_color.green;
1077 resample_filter->average_pixel.blue +=
1078 weight*resample_filter->image->background_color.blue;
1079 resample_filter->average_pixel.opacity +=
1080 resample_filter->image->background_color.opacity;
1081 divisor_c += weight;
1083 resample_filter->average_pixel.red /= divisor_c;
1084 resample_filter->average_pixel.green /= divisor_c;
1085 resample_filter->average_pixel.blue /= divisor_c;
1086 resample_filter->average_pixel.opacity /= 2;
1090 *pixel=resample_filter->average_pixel;
1097 Initialize weighted average data collection
1102 pixel->red = pixel->green = pixel->blue = 0.0;
1103 if (resample_filter->image->matte != MagickFalse) pixel->opacity = 0.0;
1104 if (resample_filter->image->colorspace == CMYKColorspace) pixel->index = 0.0;
1107 Determine the parellelogram bounding box fitted to the ellipse
1108 centered at u0,v0. This area is bounding by the lines...
1110 v1 = (ssize_t)ceil(v0 - resample_filter->Vlimit); /* range of scan lines */
1111 v2 = (ssize_t)floor(v0 + resample_filter->Vlimit);
1113 /* scan line start and width accross the parallelogram */
1114 u1 = u0 + (v1-v0)*resample_filter->slope - resample_filter->Uwidth;
1115 uw = (ssize_t)(2.0*resample_filter->Uwidth)+1;
1118 fprintf(stderr, "v1=%ld; v2=%ld\n", (long)v1, (long)v2);
1119 fprintf(stderr, "u1=%ld; uw=%ld\n", (long)u1, (long)uw);
1121 # define DEBUG_HIT_MISS 0 /* only valid if DEBUG_ELLIPSE is enabled */
1125 Do weighted resampling of all pixels, within the scaled ellipse,
1126 bound by a Parellelogram fitted to the ellipse.
1128 DDQ = 2*resample_filter->A;
1129 for( v=v1; v<=v2; v++ ) {
1131 long uu = ceil(u1); /* actual pixel location (for debug only) */
1132 fprintf(stderr, "# scan line from pixel %ld, %ld\n", (long)uu, (long)v);
1134 u = (ssize_t)ceil(u1); /* first pixel in scanline */
1135 u1 += resample_filter->slope; /* start of next scan line */
1138 /* location of this first pixel, relative to u0,v0 */
1142 /* Q = ellipse quotent ( if Q<F then pixel is inside ellipse) */
1143 Q = (resample_filter->A*U + resample_filter->B*V)*U + resample_filter->C*V*V;
1144 DQ = resample_filter->A*(2.0*U+1) + resample_filter->B*V;
1146 /* get the scanline of pixels for this v */
1147 pixels=GetCacheViewVirtualPixels(resample_filter->view,u,v,(size_t) uw,
1148 1,resample_filter->exception);
1149 if (pixels == (const PixelPacket *) NULL)
1150 return(MagickFalse);
1151 indexes=GetCacheViewVirtualIndexQueue(resample_filter->view);
1153 /* count up the weighted pixel colors */
1154 for( u=0; u<uw; u++ ) {
1156 /* Note that the ellipse has been pre-scaled so F = WLUT_WIDTH */
1157 if ( Q < (double)WLUT_WIDTH ) {
1158 weight = resample_filter->filter_lut[(int)Q];
1160 /* Note that the ellipse has been pre-scaled so F = support^2 */
1161 if ( Q < (double)resample_filter->F ) {
1162 weight = GetResizeFilterWeight(resample_filter->filter_def,
1163 sqrt(Q)); /* a SquareRoot! Arrggghhhhh... */
1166 pixel->opacity += weight*pixels->opacity;
1167 divisor_m += weight;
1169 if (resample_filter->image->matte != MagickFalse)
1170 weight *= QuantumScale*((MagickRealType)(QuantumRange-pixels->opacity));
1171 pixel->red += weight*pixels->red;
1172 pixel->green += weight*pixels->green;
1173 pixel->blue += weight*pixels->blue;
1174 if (resample_filter->image->colorspace == CMYKColorspace)
1175 pixel->index += weight*(*indexes);
1176 divisor_c += weight;
1180 /* mark the pixel according to hit/miss of the ellipse */
1181 fprintf(stderr, "set arrow from %lf,%lf to %lf,%lf nohead ls 3\n",
1182 (long)uu-.1,(double)v-.1,(long)uu+.1,(long)v+.1);
1183 fprintf(stderr, "set arrow from %lf,%lf to %lf,%lf nohead ls 3\n",
1184 (long)uu+.1,(double)v-.1,(long)uu-.1,(long)v+.1);
1186 fprintf(stderr, "set arrow from %lf,%lf to %lf,%lf nohead ls 1\n",
1187 (long)uu-.1,(double)v-.1,(long)uu+.1,(long)v+.1);
1188 fprintf(stderr, "set arrow from %lf,%lf to %lf,%lf nohead ls 1\n",
1189 (long)uu+.1,(double)v-.1,(long)uu-.1,(long)v+.1);
1202 fprintf(stderr, "Hit=%ld; Total=%ld;\n", (long)hit, (long)uw*(v2-v1) );
1206 Result sanity check -- this should NOT happen
1209 /* not enough pixels in resampling, resort to direct interpolation */
1210 #if DEBUG_NO_PIXEL_HIT
1211 pixel->opacity = pixel->red = pixel->green = pixel->blue = 0;
1212 pixel->red = QuantumRange; /* show pixels for which EWA fails */
1214 status=InterpolateResampleFilter(resample_filter,
1215 resample_filter->interpolate,u0,v0,pixel);
1221 Finialize results of resampling
1223 divisor_m = 1.0/divisor_m;
1224 pixel->opacity = (MagickRealType) ClampToQuantum(divisor_m*pixel->opacity);
1225 divisor_c = 1.0/divisor_c;
1226 pixel->red = (MagickRealType) ClampToQuantum(divisor_c*pixel->red);
1227 pixel->green = (MagickRealType) ClampToQuantum(divisor_c*pixel->green);
1228 pixel->blue = (MagickRealType) ClampToQuantum(divisor_c*pixel->blue);
1229 if (resample_filter->image->colorspace == CMYKColorspace)
1230 pixel->index = (MagickRealType) ClampToQuantum(divisor_c*pixel->index);
1234 #if EWA && EWA_CLAMP
1236 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1240 - C l a m p U p A x e s %
1244 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1246 % ClampUpAxes() function converts the input vectors into a major and
1247 % minor axis unit vectors, and their magnitude. This allows us to
1248 % ensure that the ellipse generated is never smaller than the unit
1249 % circle and thus never too small for use in EWA resampling.
1251 % This purely mathematical 'magic' was provided by Professor Nicolas
1252 % Robidoux and his Masters student Chantal Racette.
1254 % Reference: "We Recommend Singular Value Decomposition", David Austin
1255 % http://www.ams.org/samplings/feature-column/fcarc-svd
1257 % By generating major and minor axis vectors, we can actually use the
1258 % ellipse in its "canonical form", by remapping the dx,dy of the
1259 % sampled point into distances along the major and minor axis unit
1262 % Reference: http://en.wikipedia.org/wiki/Ellipse#Canonical_form
1264 static inline void ClampUpAxes(const double dux,
1270 double *major_unit_x,
1271 double *major_unit_y,
1272 double *minor_unit_x,
1273 double *minor_unit_y)
1276 * ClampUpAxes takes an input 2x2 matrix
1278 * [ a b ] = [ dux duy ]
1279 * [ c d ] = [ dvx dvy ]
1281 * and computes from it the major and minor axis vectors [major_x,
1282 * major_y] and [minor_x,minor_y] of the smallest ellipse containing
1283 * both the unit disk and the ellipse which is the image of the unit
1284 * disk by the linear transformation
1286 * [ dux duy ] [S] = [s]
1287 * [ dvx dvy ] [T] = [t]
1289 * (The vector [S,T] is the difference between a position in output
1290 * space and [X,Y]; the vector [s,t] is the difference between a
1291 * position in input space and [x,y].)
1296 * major_mag is the half-length of the major axis of the "new"
1299 * minor_mag is the half-length of the minor axis of the "new"
1302 * major_unit_x is the x-coordinate of the major axis direction vector
1303 * of both the "old" and "new" ellipses.
1305 * major_unit_y is the y-coordinate of the major axis direction vector.
1307 * minor_unit_x is the x-coordinate of the minor axis direction vector.
1309 * minor_unit_y is the y-coordinate of the minor axis direction vector.
1311 * Unit vectors are useful for computing projections, in particular,
1312 * to compute the distance between a point in output space and the
1313 * center (of a disk) from the position of the corresponding point
1316 * Now, if you want to modify the input pair of tangent vectors so
1317 * that it defines the modified ellipse, all you have to do is set
1319 * newdux = major_mag * major_unit_x
1320 * newdvx = major_mag * major_unit_y
1321 * newduy = minor_mag * minor_unit_x = minor_mag * -major_unit_y
1322 * newdvy = minor_mag * minor_unit_y = minor_mag * major_unit_x
1324 * and use these tangent vectors as if they were the original ones.
1325 * Usually, this is a drastic change in the tangent vectors even if
1326 * the singular values are not clamped; for example, the minor axis
1327 * vector always points in a direction which is 90 degrees
1328 * counterclockwise from the direction of the major axis vector.
1333 * GOAL: Fix things so that the pullback, in input space, of a disk
1334 * of radius r in output space is an ellipse which contains, at
1335 * least, a disc of radius r. (Make this hold for any r>0.)
1337 * ESSENCE OF THE METHOD: Compute the product of the first two
1338 * factors of an SVD of the linear transformation defining the
1339 * ellipse and make sure that both its columns have norm at least 1.
1340 * Because rotations and reflexions map disks to themselves, it is
1341 * not necessary to compute the third (rightmost) factor of the SVD.
1343 * DETAILS: Find the singular values and (unit) left singular
1344 * vectors of Jinv, clampling up the singular values to 1, and
1345 * multiply the unit left singular vectors by the new singular
1346 * values in order to get the minor and major ellipse axis vectors.
1348 * Image resampling context:
1350 * The Jacobian matrix of the transformation at the output point
1351 * under consideration is defined as follows:
1353 * Consider the transformation (x,y) -> (X,Y) from input locations
1354 * to output locations. (Anthony Thyssen, elsewhere in resample.c,
1355 * uses the notation (u,v) -> (x,y).)
1357 * The Jacobian matrix of the transformation at (x,y) is equal to
1359 * J = [ A, B ] = [ dX/dx, dX/dy ]
1360 * [ C, D ] [ dY/dx, dY/dy ]
1362 * that is, the vector [A,C] is the tangent vector corresponding to
1363 * input changes in the horizontal direction, and the vector [B,D]
1364 * is the tangent vector corresponding to input changes in the
1365 * vertical direction.
1367 * In the context of resampling, it is natural to use the inverse
1368 * Jacobian matrix Jinv because resampling is generally performed by
1369 * pulling pixel locations in the output image back to locations in
1370 * the input image. Jinv is
1372 * Jinb = [ a, b ] = [ dx/dX, dx/dY ]
1373 * [ c, d ] [ dy/dX, dy/dY ]
1375 * Note: Jinv can be computed from J with the following matrix
1378 * Jinv = 1/(A*D-B*C) [ D, -B ]
1381 * What we do is modify Jinv so that it generates an ellipse which
1382 * is as close as possible to the original but which contains the
1383 * unit disk. This can be accomplished as follows:
1387 * Jinv = U Sigma V^T
1389 * be an SVD decomposition of Jinv. (The SVD is not unique, but the
1390 * final ellipse does not depend on the particular SVD.)
1392 * We could clamp up the entries of the diagonal matrix Sigma so
1393 * that they are at least 1, and then set
1395 * Jinv = U newSigma V^T.
1397 * However, we do not need to compute V for the following reason:
1398 * V^T is an orthogonal matrix (that is, it represents a combination
1399 * of rotations and reflexions) so that it maps the unit circle to
1400 * itself. For this reason, the exact value of V does not affect the
1401 * final ellipse, and we can choose V to be the identity
1402 * matrix. This gives
1404 * Jinv = U newSigma.
1406 * In the end, we return the two diagonal entries of newSigma
1407 * together with the two columns of U.
1410 * ClampUpAxes was written by Nicolas Robidoux and Chantal Racette
1411 * of Laurentian University with funding from the National Science
1412 * and Engineering Research Council of Canada.
1414 * The astrophysicist Craig DeForest pioneered the use of the SVD to
1415 * clamp up the singular values of the Jacobian matrix of the
1416 * pullback transformation for EWA resampling. It is implemented in
1417 * his PDL::Transform code (PDL = Perl Data Language).
1419 const double a = dux;
1420 const double b = duy;
1421 const double c = dvx;
1422 const double d = dvy;
1424 * n is the matrix Jinv * transpose(Jinv). Eigenvalues of n are the
1425 * squares of the singular values of Jinv.
1427 const double aa = a*a;
1428 const double bb = b*b;
1429 const double cc = c*c;
1430 const double dd = d*d;
1432 * Eigenvectors of n are left singular vectors of Jinv.
1434 const double n11 = aa+bb;
1435 const double n12 = a*c+b*d;
1436 const double n21 = n12;
1437 const double n22 = cc+dd;
1438 const double det = a*d-b*c;
1439 const double twice_det = det+det;
1440 const double frobenius_squared = n11+n22;
1441 const double discriminant =
1442 (frobenius_squared+twice_det)*(frobenius_squared-twice_det);
1443 const double sqrt_discriminant = sqrt(discriminant);
1445 * s1 is the largest singular value of the inverse Jacobian
1446 * matrix. In other words, its reciprocal is the smallest singular
1447 * value of the Jacobian matrix itself.
1448 * If s1 = 0, both singular values are 0, and any orthogonal pair of
1449 * left and right factors produces a singular decomposition of Jinv.
1452 * Initially, we only compute the squares of the singular values.
1454 const double s1s1 = 0.5*(frobenius_squared+sqrt_discriminant);
1456 * s2 the smallest singular value of the inverse Jacobian
1457 * matrix. Its reciprocal is the largest singular value of the
1458 * Jacobian matrix itself.
1460 const double s2s2 = 0.5*(frobenius_squared-sqrt_discriminant);
1461 const double s1s1minusn11 = s1s1-n11;
1462 const double s1s1minusn22 = s1s1-n22;
1464 * u1, the first column of the U factor of a singular decomposition
1465 * of Jinv, is a (non-normalized) left singular vector corresponding
1466 * to s1. It has entries u11 and u21. We compute u1 from the fact
1467 * that it is an eigenvector of n corresponding to the eigenvalue
1470 const double s1s1minusn11_squared = s1s1minusn11*s1s1minusn11;
1471 const double s1s1minusn22_squared = s1s1minusn22*s1s1minusn22;
1473 * The following selects the largest row of n-s1^2 I as the one
1474 * which is used to find the eigenvector. If both s1^2-n11 and
1475 * s1^2-n22 are zero, n-s1^2 I is the zero matrix. In that case,
1476 * any vector is an eigenvector; in addition, norm below is equal to
1477 * zero, and, in exact arithmetic, this is the only case in which
1478 * norm = 0. So, setting u1 to the simple but arbitrary vector [1,0]
1479 * if norm = 0 safely takes care of all cases.
1481 const double temp_u11 =
1482 ( (s1s1minusn11_squared>=s1s1minusn22_squared) ? n12 : s1s1minusn22 );
1483 const double temp_u21 =
1484 ( (s1s1minusn11_squared>=s1s1minusn22_squared) ? s1s1minusn11 : n21 );
1485 const double norm = sqrt(temp_u11*temp_u11+temp_u21*temp_u21);
1487 * Finalize the entries of first left singular vector (associated
1488 * with the largest singular value).
1490 const double u11 = ( (norm>0.0) ? temp_u11/norm : 1.0 );
1491 const double u21 = ( (norm>0.0) ? temp_u21/norm : 0.0 );
1493 * Clamp the singular values up to 1.
1495 *major_mag = ( (s1s1<=1.0) ? 1.0 : sqrt(s1s1) );
1496 *minor_mag = ( (s2s2<=1.0) ? 1.0 : sqrt(s2s2) );
1498 * Return the unit major and minor axis direction vectors.
1500 *major_unit_x = u11;
1501 *major_unit_y = u21;
1502 *minor_unit_x = -u21;
1503 *minor_unit_y = u11;
1508 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1512 % S c a l e R e s a m p l e F i l t e r %
1516 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1518 % ScaleResampleFilter() does all the calculations needed to resample an image
1519 % at a specific scale, defined by two scaling vectors. This not using
1520 % a orthogonal scaling, but two distorted scaling vectors, to allow the
1521 % generation of a angled ellipse.
1523 % As only two deritive scaling vectors are used the center of the ellipse
1524 % must be the center of the lookup. That is any curvature that the
1525 % distortion may produce is discounted.
1527 % The input vectors are produced by either finding the derivitives of the
1528 % distortion function, or the partial derivitives from a distortion mapping.
1529 % They do not need to be the orthogonal dx,dy scaling vectors, but can be
1530 % calculated from other derivatives. For example you could use dr,da/r
1531 % polar coordinate vector scaling vectors
1533 % If u,v = DistortEquation(x,y) OR u = Fu(x,y); v = Fv(x,y)
1534 % Then the scaling vectors are determined from the deritives...
1535 % du/dx, dv/dx and du/dy, dv/dy
1536 % If the resulting scaling vectors is othogonally aligned then...
1537 % dv/dx = 0 and du/dy = 0
1538 % Producing an othogonally alligned ellipse in source space for the area to
1541 % Note that scaling vectors are different to argument order. Argument order
1542 % is the general order the deritives are extracted from the distortion
1543 % equations, and not the scaling vectors. As such the middle two vaules
1544 % may be swapped from what you expect. Caution is advised.
1546 % WARNING: It is assumed that any SetResampleFilter() method call will
1547 % always be performed before the ScaleResampleFilter() method, so that the
1548 % size of the ellipse will match the support for the resampling filter being
1551 % The format of the ScaleResampleFilter method is:
1553 % void ScaleResampleFilter(const ResampleFilter *resample_filter,
1554 % const double dux,const double duy,const double dvx,const double dvy)
1556 % A description of each parameter follows:
1558 % o resample_filter: the resampling resample_filterrmation defining the
1559 % image being resampled
1561 % o dux,duy,dvx,dvy:
1562 % The deritives or scaling vectors defining the EWA ellipse.
1563 % NOTE: watch the order, which is based on the order deritives
1564 % are usally determined from distortion equations (see above).
1565 % The middle two values may need to be swapped if you are thinking
1566 % in terms of scaling vectors.
1569 MagickExport void ScaleResampleFilter(ResampleFilter *resample_filter,
1570 const double dux,const double duy,const double dvx,const double dvy)
1574 assert(resample_filter != (ResampleFilter *) NULL);
1575 assert(resample_filter->signature == MagickSignature);
1577 resample_filter->limit_reached = MagickFalse;
1579 /* A 'point' filter forces use of interpolation instead of area sampling */
1580 if ( resample_filter->filter == PointFilter )
1581 return; /* EWA turned off - nothing to do */
1584 fprintf(stderr, "# -----\n" );
1585 fprintf(stderr, "dux=%lf; dvx=%lf; duy=%lf; dvy=%lf;\n",
1586 dux, dvx, duy, dvy);
1589 /* Find Ellipse Coefficents such that
1590 A*u^2 + B*u*v + C*v^2 = F
1591 With u,v relative to point around which we are resampling.
1592 And the given scaling dx,dy vectors in u,v space
1593 du/dx,dv/dx and du/dy,dv/dy
1596 /* Direct conversion of derivatives into elliptical coefficients
1597 However when magnifying images, the scaling vectors will be small
1598 resulting in a ellipse that is too small to sample properly.
1599 As such we need to clamp the major/minor axis to a minumum of 1.0
1600 to prevent it getting too small.
1610 ClampUpAxes(dux,dvx,duy,dvy, &major_mag, &minor_mag,
1611 &major_x, &major_y, &minor_x, &minor_y);
1612 major_x *= major_mag; major_y *= major_mag;
1613 minor_x *= minor_mag; minor_y *= minor_mag;
1615 fprintf(stderr, "major_x=%lf; major_y=%lf; minor_x=%lf; minor_y=%lf;\n",
1616 major_x, major_y, minor_x, minor_y);
1618 A = major_y*major_y+minor_y*minor_y;
1619 B = -2.0*(major_x*major_y+minor_x*minor_y);
1620 C = major_x*major_x+minor_x*minor_x;
1621 F = major_mag*minor_mag;
1622 F *= F; /* square it */
1624 #else /* raw unclamped EWA */
1625 A = dvx*dvx+dvy*dvy;
1626 B = -2.0*(dux*dvx+duy*dvy);
1627 C = dux*dux+duy*duy;
1628 F = dux*dvy-duy*dvx;
1629 F *= F; /* square it */
1630 #endif /* EWA_CLAMP */
1634 This Paul Heckbert's "Higher Quality EWA" formula, from page 60 in his
1635 thesis, which adds a unit circle to the elliptical area so as to do both
1636 Reconstruction and Prefiltering of the pixels in the resampling. It also
1637 means it is always likely to have at least 4 pixels within the area of the
1638 ellipse, for weighted averaging. No scaling will result with F == 4.0 and
1639 a circle of radius 2.0, and F smaller than this means magnification is
1642 NOTE: This method produces a very blury result at near unity scale while
1643 producing perfect results for strong minitification and magnifications.
1645 However filter support is fixed to 2.0 (no good for Windowed Sinc filters)
1647 A = dvx*dvx+dvy*dvy+1;
1648 B = -2.0*(dux*dvx+duy*dvy);
1649 C = dux*dux+duy*duy+1;
1654 fprintf(stderr, "A=%lf; B=%lf; C=%lf; F=%lf\n", A,B,C,F);
1656 /* Figure out the various information directly about the ellipse.
1657 This information currently not needed at this time, but may be
1658 needed later for better limit determination.
1660 It is also good to have as a record for future debugging
1662 { double alpha, beta, gamma, Major, Minor;
1663 double Eccentricity, Ellipse_Area, Ellipse_Angle;
1667 gamma = sqrt(beta*beta + B*B );
1669 if ( alpha - gamma <= MagickEpsilon )
1672 Major = sqrt(2*F/(alpha - gamma));
1673 Minor = sqrt(2*F/(alpha + gamma));
1675 fprintf(stderr, "# Major=%lf; Minor=%lf\n", Major, Minor );
1677 /* other information about ellipse include... */
1678 Eccentricity = Major/Minor;
1679 Ellipse_Area = MagickPI*Major*Minor;
1680 Ellipse_Angle = atan2(B, A-C);
1682 fprintf(stderr, "# Angle=%lf Area=%lf\n",
1683 RadiansToDegrees(Ellipse_Angle), Ellipse_Area);
1687 /* If one or both of the scaling vectors is impossibly large
1688 (producing a very large raw F value), we may as well not bother
1689 doing any form of resampling since resampled area is very large.
1690 In this case some alternative means of pixel sampling, such as
1691 the average of the whole image is needed to get a reasonable
1692 result. Calculate only as needed.
1694 if ( (4*A*C - B*B) > MagickHuge ) {
1695 resample_filter->limit_reached = MagickTrue;
1699 /* Scale ellipse to match the filters support
1700 (that is, multiply F by the square of the support).
1702 F *= resample_filter->support;
1703 F *= resample_filter->support;
1705 /* Orthogonal bounds of the ellipse */
1706 resample_filter->Ulimit = sqrt(4*C*F/(4*A*C-B*B));
1707 resample_filter->Vlimit = sqrt(4*A*F/(4*A*C-B*B));
1709 /* Horizontally aligned parallelogram fitted to Ellipse */
1710 resample_filter->Uwidth = sqrt(F/A); /* Half of the parallelogram width */
1711 resample_filter->slope = -B/(2*A); /* Reciprocal slope of the parallelogram */
1714 fprintf(stderr, "Ulimit=%lf; Vlimit=%lf; UWidth=%lf; Slope=%lf;\n",
1715 resample_filter->Ulimit, resample_filter->Vlimit,
1716 resample_filter->Uwidth, resample_filter->slope );
1719 /* Check the absolute area of the parallelogram involved.
1720 * This limit needs more work, as it is too slow for larger images
1721 * with tiled views of the horizon.
1723 if ( (resample_filter->Uwidth * resample_filter->Vlimit)
1724 > (4.0*resample_filter->image_area)) {
1725 resample_filter->limit_reached = MagickTrue;
1729 /* Scale ellipse formula to directly index the Filter Lookup Table */
1730 { register double scale;
1732 /* scale so that F = WLUT_WIDTH; -- hardcoded */
1733 scale = (double)WLUT_WIDTH/F;
1735 /* scale so that F = resample_filter->F (support^2) */
1736 scale = resample_filter->F/F;
1738 resample_filter->A = A*scale;
1739 resample_filter->B = B*scale;
1740 resample_filter->C = C*scale;
1745 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1749 % S e t R e s a m p l e F i l t e r %
1753 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1755 % SetResampleFilter() set the resampling filter lookup table based on a
1756 % specific filter. Note that the filter is used as a radial filter not as a
1757 % two pass othogonally aligned resampling filter.
1759 % The default Filter, is Gaussian, which is the standard filter used by the
1760 % original paper on the Elliptical Weighted Everage Algorithm. However other
1761 % filters can also be used.
1763 % The format of the SetResampleFilter method is:
1765 % void SetResampleFilter(ResampleFilter *resample_filter,
1766 % const FilterTypes filter,const double blur)
1768 % A description of each parameter follows:
1770 % o resample_filter: resampling resample_filterrmation structure
1772 % o filter: the resize filter for elliptical weighting LUT
1774 % o blur: filter blur factor (radial scaling) for elliptical weighting LUT
1777 MagickExport void SetResampleFilter(ResampleFilter *resample_filter,
1778 const FilterTypes filter,const double blur)
1783 assert(resample_filter != (ResampleFilter *) NULL);
1784 assert(resample_filter->signature == MagickSignature);
1786 resample_filter->do_interpolate = MagickFalse;
1787 resample_filter->filter = filter;
1789 if ( filter == PointFilter )
1791 resample_filter->do_interpolate = MagickTrue;
1792 return; /* EWA turned off - nothing more to do */
1795 /* Set a default cylindrical filter of a 'low blur' Jinc windowed Jinc */
1796 if ( filter == UndefinedFilter )
1797 resample_filter->filter = RobidouxFilter;
1799 resize_filter = AcquireResizeFilter(resample_filter->image,
1800 resample_filter->filter,blur,MagickTrue,resample_filter->exception);
1801 if (resize_filter == (ResizeFilter *) NULL)
1803 (void) ThrowMagickException(resample_filter->exception,GetMagickModule(),
1804 ModuleError, "UnableToSetFilteringValue",
1805 "Fall back to default EWA gaussian filter");
1806 resample_filter->filter = PointFilter;
1809 /* Get the practical working support for the filter,
1810 * after any API call blur factors have been accoded for.
1813 resample_filter->support = GetResizeFilterSupport(resize_filter);
1815 resample_filter->support = 2.0; /* fixed support size for HQ-EWA */
1819 /* Fill the LUT with the weights from the selected filter function */
1824 /* Scale radius so the filter LUT covers the full support range */
1825 r_scale = resample_filter->support*sqrt(1.0/(double)WLUT_WIDTH);
1826 for(Q=0; Q<WLUT_WIDTH; Q++)
1827 resample_filter->filter_lut[Q] = (double)
1828 GetResizeFilterWeight(resize_filter,sqrt((double)Q)*r_scale);
1830 /* finished with the resize filter */
1831 resize_filter = DestroyResizeFilter(resize_filter);
1834 /* save the filter and the scaled ellipse bounds needed for filter */
1835 resample_filter->filter_def = resize_filter;
1836 resample_filter->F = resample_filter->support*resample_filter->support;
1840 Adjust the scaling of the default unit circle
1841 This assumes that any real scaling changes will always
1842 take place AFTER the filter method has been initialized.
1844 ScaleResampleFilter(resample_filter, 1.0, 0.0, 0.0, 1.0);
1847 /* This is old code kept as a reference only. It is very wrong,
1848 and I don't understand exactly what it was attempting to do.
1851 Create Normal Gaussian 2D Filter Weighted Lookup Table.
1852 A normal EWA guassual lookup would use exp(Q*ALPHA)
1853 where Q = distance squared from 0.0 (center) to 1.0 (edge)
1854 and ALPHA = -4.0*ln(2.0) ==> -2.77258872223978123767
1855 The table is of length 1024, and equates to support radius of 2.0
1856 thus needs to be scaled by ALPHA*4/1024 and any blur factor squared
1858 The above came from some reference code provided by Fred Weinhaus
1859 and seems to have been a guess that was appropriate for its use
1860 in a 3d perspective landscape mapping program.
1862 r_scale = -2.77258872223978123767/(WLUT_WIDTH*blur*blur);
1863 for(Q=0; Q<WLUT_WIDTH; Q++)
1864 resample_filter->filter_lut[Q] = exp((double)Q*r_scale);
1865 resample_filter->support = WLUT_WIDTH;
1870 #if defined(MAGICKCORE_OPENMP_SUPPORT)
1878 /* Scale radius so the filter LUT covers the full support range */
1879 r_scale = resample_filter->support*sqrt(1.0/(double)WLUT_WIDTH);
1880 if (IsMagickTrue(GetImageArtifact(resample_filter->image,"resample:verbose")) )
1882 /* Debug output of the filter weighting LUT
1883 Gnuplot the LUT with hoizontal adjusted to 'r' using...
1884 plot [0:2][-.2:1] "lut.dat" using (sqrt($0/1024)*2):1 with lines
1885 The filter values is normalized for comparision
1888 printf("# Resampling Filter LUT (%d values)\n", WLUT_WIDTH);
1890 printf("# Note: values in table are using a squared radius lookup.\n");
1891 printf("# And the whole table represents the filters support.\n");
1892 printf("\n"); /* generates a 'break' in gnuplot if multiple outputs */
1893 for(Q=0; Q<WLUT_WIDTH; Q++)
1894 printf("%8.*g %.*g\n",
1895 GetMagickPrecision(),sqrt((double)Q)*r_scale,
1896 GetMagickPrecision(),resample_filter->filter_lut[Q] );
1898 /* output the above once only for each image, and each setting */
1899 (void) DeleteImageArtifact(resample_filter->image,"resample:verbose");
1901 #endif /* FILTER_LUT */
1906 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1910 % S e t R e s a m p l e F i l t e r I n t e r p o l a t e M e t h o d %
1914 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1916 % SetResampleFilterInterpolateMethod() changes the interpolation method
1917 % associated with the specified resample filter.
1919 % The format of the SetResampleFilterInterpolateMethod method is:
1921 % MagickBooleanType SetResampleFilterInterpolateMethod(
1922 % ResampleFilter *resample_filter,const InterpolateMethod method)
1924 % A description of each parameter follows:
1926 % o resample_filter: the resample filter.
1928 % o method: the interpolation method.
1931 MagickExport MagickBooleanType SetResampleFilterInterpolateMethod(
1932 ResampleFilter *resample_filter,const InterpolatePixelMethod method)
1934 assert(resample_filter != (ResampleFilter *) NULL);
1935 assert(resample_filter->signature == MagickSignature);
1936 assert(resample_filter->image != (Image *) NULL);
1938 if (resample_filter->debug != MagickFalse)
1939 (void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",
1940 resample_filter->image->filename);
1942 resample_filter->interpolate=method;
1948 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1952 % S e t R e s a m p l e F i l t e r V i r t u a l P i x e l M e t h o d %
1956 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1958 % SetResampleFilterVirtualPixelMethod() changes the virtual pixel method
1959 % associated with the specified resample filter.
1961 % The format of the SetResampleFilterVirtualPixelMethod method is:
1963 % MagickBooleanType SetResampleFilterVirtualPixelMethod(
1964 % ResampleFilter *resample_filter,const VirtualPixelMethod method)
1966 % A description of each parameter follows:
1968 % o resample_filter: the resample filter.
1970 % o method: the virtual pixel method.
1973 MagickExport MagickBooleanType SetResampleFilterVirtualPixelMethod(
1974 ResampleFilter *resample_filter,const VirtualPixelMethod method)
1976 assert(resample_filter != (ResampleFilter *) NULL);
1977 assert(resample_filter->signature == MagickSignature);
1978 assert(resample_filter->image != (Image *) NULL);
1979 if (resample_filter->debug != MagickFalse)
1980 (void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",
1981 resample_filter->image->filename);
1982 resample_filter->virtual_pixel=method;
1983 if (method != UndefinedVirtualPixelMethod)
1984 (void) SetCacheViewVirtualPixelMethod(resample_filter->view,method);