Line data Source code
1 : /*
2 : * jquant2.c
3 : *
4 : * This file was part of the Independent JPEG Group's software:
5 : * Copyright (C) 1991-1996, Thomas G. Lane.
6 : * libjpeg-turbo Modifications:
7 : * Copyright (C) 2009, 2014-2015, D. R. Commander.
8 : * For conditions of distribution and use, see the accompanying README.ijg
9 : * file.
10 : *
11 : * This file contains 2-pass color quantization (color mapping) routines.
12 : * These routines provide selection of a custom color map for an image,
13 : * followed by mapping of the image to that color map, with optional
14 : * Floyd-Steinberg dithering.
15 : * It is also possible to use just the second pass to map to an arbitrary
16 : * externally-given color map.
17 : *
18 : * Note: ordered dithering is not supported, since there isn't any fast
19 : * way to compute intercolor distances; it's unclear that ordered dither's
20 : * fundamental assumptions even hold with an irregularly spaced color map.
21 : */
22 :
23 : #define JPEG_INTERNALS
24 : #include "jinclude.h"
25 : #include "jpeglib.h"
26 :
27 : #ifdef QUANT_2PASS_SUPPORTED
28 :
29 :
30 : /*
31 : * This module implements the well-known Heckbert paradigm for color
32 : * quantization. Most of the ideas used here can be traced back to
33 : * Heckbert's seminal paper
34 : * Heckbert, Paul. "Color Image Quantization for Frame Buffer Display",
35 : * Proc. SIGGRAPH '82, Computer Graphics v.16 #3 (July 1982), pp 297-304.
36 : *
37 : * In the first pass over the image, we accumulate a histogram showing the
38 : * usage count of each possible color. To keep the histogram to a reasonable
39 : * size, we reduce the precision of the input; typical practice is to retain
40 : * 5 or 6 bits per color, so that 8 or 4 different input values are counted
41 : * in the same histogram cell.
42 : *
43 : * Next, the color-selection step begins with a box representing the whole
44 : * color space, and repeatedly splits the "largest" remaining box until we
45 : * have as many boxes as desired colors. Then the mean color in each
46 : * remaining box becomes one of the possible output colors.
47 : *
48 : * The second pass over the image maps each input pixel to the closest output
49 : * color (optionally after applying a Floyd-Steinberg dithering correction).
50 : * This mapping is logically trivial, but making it go fast enough requires
51 : * considerable care.
52 : *
53 : * Heckbert-style quantizers vary a good deal in their policies for choosing
54 : * the "largest" box and deciding where to cut it. The particular policies
55 : * used here have proved out well in experimental comparisons, but better ones
56 : * may yet be found.
57 : *
58 : * In earlier versions of the IJG code, this module quantized in YCbCr color
59 : * space, processing the raw upsampled data without a color conversion step.
60 : * This allowed the color conversion math to be done only once per colormap
61 : * entry, not once per pixel. However, that optimization precluded other
62 : * useful optimizations (such as merging color conversion with upsampling)
63 : * and it also interfered with desired capabilities such as quantizing to an
64 : * externally-supplied colormap. We have therefore abandoned that approach.
65 : * The present code works in the post-conversion color space, typically RGB.
66 : *
67 : * To improve the visual quality of the results, we actually work in scaled
68 : * RGB space, giving G distances more weight than R, and R in turn more than
69 : * B. To do everything in integer math, we must use integer scale factors.
70 : * The 2/3/1 scale factors used here correspond loosely to the relative
71 : * weights of the colors in the NTSC grayscale equation.
72 : * If you want to use this code to quantize a non-RGB color space, you'll
73 : * probably need to change these scale factors.
74 : */
75 :
76 : #define R_SCALE 2 /* scale R distances by this much */
77 : #define G_SCALE 3 /* scale G distances by this much */
78 : #define B_SCALE 1 /* and B by this much */
79 :
80 : static const int c_scales[3]={R_SCALE, G_SCALE, B_SCALE};
81 : #define C0_SCALE c_scales[rgb_red[cinfo->out_color_space]]
82 : #define C1_SCALE c_scales[rgb_green[cinfo->out_color_space]]
83 : #define C2_SCALE c_scales[rgb_blue[cinfo->out_color_space]]
84 :
85 : /*
86 : * First we have the histogram data structure and routines for creating it.
87 : *
88 : * The number of bits of precision can be adjusted by changing these symbols.
89 : * We recommend keeping 6 bits for G and 5 each for R and B.
90 : * If you have plenty of memory and cycles, 6 bits all around gives marginally
91 : * better results; if you are short of memory, 5 bits all around will save
92 : * some space but degrade the results.
93 : * To maintain a fully accurate histogram, we'd need to allocate a "long"
94 : * (preferably unsigned long) for each cell. In practice this is overkill;
95 : * we can get by with 16 bits per cell. Few of the cell counts will overflow,
96 : * and clamping those that do overflow to the maximum value will give close-
97 : * enough results. This reduces the recommended histogram size from 256Kb
98 : * to 128Kb, which is a useful savings on PC-class machines.
99 : * (In the second pass the histogram space is re-used for pixel mapping data;
100 : * in that capacity, each cell must be able to store zero to the number of
101 : * desired colors. 16 bits/cell is plenty for that too.)
102 : * Since the JPEG code is intended to run in small memory model on 80x86
103 : * machines, we can't just allocate the histogram in one chunk. Instead
104 : * of a true 3-D array, we use a row of pointers to 2-D arrays. Each
105 : * pointer corresponds to a C0 value (typically 2^5 = 32 pointers) and
106 : * each 2-D array has 2^6*2^5 = 2048 or 2^6*2^6 = 4096 entries.
107 : */
108 :
109 : #define MAXNUMCOLORS (MAXJSAMPLE+1) /* maximum size of colormap */
110 :
111 : /* These will do the right thing for either R,G,B or B,G,R color order,
112 : * but you may not like the results for other color orders.
113 : */
114 : #define HIST_C0_BITS 5 /* bits of precision in R/B histogram */
115 : #define HIST_C1_BITS 6 /* bits of precision in G histogram */
116 : #define HIST_C2_BITS 5 /* bits of precision in B/R histogram */
117 :
118 : /* Number of elements along histogram axes. */
119 : #define HIST_C0_ELEMS (1<<HIST_C0_BITS)
120 : #define HIST_C1_ELEMS (1<<HIST_C1_BITS)
121 : #define HIST_C2_ELEMS (1<<HIST_C2_BITS)
122 :
123 : /* These are the amounts to shift an input value to get a histogram index. */
124 : #define C0_SHIFT (BITS_IN_JSAMPLE-HIST_C0_BITS)
125 : #define C1_SHIFT (BITS_IN_JSAMPLE-HIST_C1_BITS)
126 : #define C2_SHIFT (BITS_IN_JSAMPLE-HIST_C2_BITS)
127 :
128 :
129 : typedef UINT16 histcell; /* histogram cell; prefer an unsigned type */
130 :
131 : typedef histcell *histptr; /* for pointers to histogram cells */
132 :
133 : typedef histcell hist1d[HIST_C2_ELEMS]; /* typedefs for the array */
134 : typedef hist1d *hist2d; /* type for the 2nd-level pointers */
135 : typedef hist2d *hist3d; /* type for top-level pointer */
136 :
137 :
138 : /* Declarations for Floyd-Steinberg dithering.
139 : *
140 : * Errors are accumulated into the array fserrors[], at a resolution of
141 : * 1/16th of a pixel count. The error at a given pixel is propagated
142 : * to its not-yet-processed neighbors using the standard F-S fractions,
143 : * ... (here) 7/16
144 : * 3/16 5/16 1/16
145 : * We work left-to-right on even rows, right-to-left on odd rows.
146 : *
147 : * We can get away with a single array (holding one row's worth of errors)
148 : * by using it to store the current row's errors at pixel columns not yet
149 : * processed, but the next row's errors at columns already processed. We
150 : * need only a few extra variables to hold the errors immediately around the
151 : * current column. (If we are lucky, those variables are in registers, but
152 : * even if not, they're probably cheaper to access than array elements are.)
153 : *
154 : * The fserrors[] array has (#columns + 2) entries; the extra entry at
155 : * each end saves us from special-casing the first and last pixels.
156 : * Each entry is three values long, one value for each color component.
157 : */
158 :
159 : #if BITS_IN_JSAMPLE == 8
160 : typedef INT16 FSERROR; /* 16 bits should be enough */
161 : typedef int LOCFSERROR; /* use 'int' for calculation temps */
162 : #else
163 : typedef JLONG FSERROR; /* may need more than 16 bits */
164 : typedef JLONG LOCFSERROR; /* be sure calculation temps are big enough */
165 : #endif
166 :
167 : typedef FSERROR *FSERRPTR; /* pointer to error array */
168 :
169 :
170 : /* Private subobject */
171 :
172 : typedef struct {
173 : struct jpeg_color_quantizer pub; /* public fields */
174 :
175 : /* Space for the eventually created colormap is stashed here */
176 : JSAMPARRAY sv_colormap; /* colormap allocated at init time */
177 : int desired; /* desired # of colors = size of colormap */
178 :
179 : /* Variables for accumulating image statistics */
180 : hist3d histogram; /* pointer to the histogram */
181 :
182 : boolean needs_zeroed; /* TRUE if next pass must zero histogram */
183 :
184 : /* Variables for Floyd-Steinberg dithering */
185 : FSERRPTR fserrors; /* accumulated errors */
186 : boolean on_odd_row; /* flag to remember which row we are on */
187 : int *error_limiter; /* table for clamping the applied error */
188 : } my_cquantizer;
189 :
190 : typedef my_cquantizer *my_cquantize_ptr;
191 :
192 :
193 : /*
194 : * Prescan some rows of pixels.
195 : * In this module the prescan simply updates the histogram, which has been
196 : * initialized to zeroes by start_pass.
197 : * An output_buf parameter is required by the method signature, but no data
198 : * is actually output (in fact the buffer controller is probably passing a
199 : * NULL pointer).
200 : */
201 :
202 : METHODDEF(void)
203 0 : prescan_quantize (j_decompress_ptr cinfo, JSAMPARRAY input_buf,
204 : JSAMPARRAY output_buf, int num_rows)
205 : {
206 0 : my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
207 : register JSAMPROW ptr;
208 : register histptr histp;
209 0 : register hist3d histogram = cquantize->histogram;
210 : int row;
211 : JDIMENSION col;
212 0 : JDIMENSION width = cinfo->output_width;
213 :
214 0 : for (row = 0; row < num_rows; row++) {
215 0 : ptr = input_buf[row];
216 0 : for (col = width; col > 0; col--) {
217 : /* get pixel value and index into the histogram */
218 0 : histp = & histogram[GETJSAMPLE(ptr[0]) >> C0_SHIFT]
219 0 : [GETJSAMPLE(ptr[1]) >> C1_SHIFT]
220 0 : [GETJSAMPLE(ptr[2]) >> C2_SHIFT];
221 : /* increment, check for overflow and undo increment if so. */
222 0 : if (++(*histp) <= 0)
223 0 : (*histp)--;
224 0 : ptr += 3;
225 : }
226 : }
227 0 : }
228 :
229 :
230 : /*
231 : * Next we have the really interesting routines: selection of a colormap
232 : * given the completed histogram.
233 : * These routines work with a list of "boxes", each representing a rectangular
234 : * subset of the input color space (to histogram precision).
235 : */
236 :
237 : typedef struct {
238 : /* The bounds of the box (inclusive); expressed as histogram indexes */
239 : int c0min, c0max;
240 : int c1min, c1max;
241 : int c2min, c2max;
242 : /* The volume (actually 2-norm) of the box */
243 : JLONG volume;
244 : /* The number of nonzero histogram cells within this box */
245 : long colorcount;
246 : } box;
247 :
248 : typedef box *boxptr;
249 :
250 :
251 : LOCAL(boxptr)
252 0 : find_biggest_color_pop (boxptr boxlist, int numboxes)
253 : /* Find the splittable box with the largest color population */
254 : /* Returns NULL if no splittable boxes remain */
255 : {
256 : register boxptr boxp;
257 : register int i;
258 0 : register long maxc = 0;
259 0 : boxptr which = NULL;
260 :
261 0 : for (i = 0, boxp = boxlist; i < numboxes; i++, boxp++) {
262 0 : if (boxp->colorcount > maxc && boxp->volume > 0) {
263 0 : which = boxp;
264 0 : maxc = boxp->colorcount;
265 : }
266 : }
267 0 : return which;
268 : }
269 :
270 :
271 : LOCAL(boxptr)
272 0 : find_biggest_volume (boxptr boxlist, int numboxes)
273 : /* Find the splittable box with the largest (scaled) volume */
274 : /* Returns NULL if no splittable boxes remain */
275 : {
276 : register boxptr boxp;
277 : register int i;
278 0 : register JLONG maxv = 0;
279 0 : boxptr which = NULL;
280 :
281 0 : for (i = 0, boxp = boxlist; i < numboxes; i++, boxp++) {
282 0 : if (boxp->volume > maxv) {
283 0 : which = boxp;
284 0 : maxv = boxp->volume;
285 : }
286 : }
287 0 : return which;
288 : }
289 :
290 :
291 : LOCAL(void)
292 0 : update_box (j_decompress_ptr cinfo, boxptr boxp)
293 : /* Shrink the min/max bounds of a box to enclose only nonzero elements, */
294 : /* and recompute its volume and population */
295 : {
296 0 : my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
297 0 : hist3d histogram = cquantize->histogram;
298 : histptr histp;
299 : int c0,c1,c2;
300 : int c0min,c0max,c1min,c1max,c2min,c2max;
301 : JLONG dist0,dist1,dist2;
302 : long ccount;
303 :
304 0 : c0min = boxp->c0min; c0max = boxp->c0max;
305 0 : c1min = boxp->c1min; c1max = boxp->c1max;
306 0 : c2min = boxp->c2min; c2max = boxp->c2max;
307 :
308 0 : if (c0max > c0min)
309 0 : for (c0 = c0min; c0 <= c0max; c0++)
310 0 : for (c1 = c1min; c1 <= c1max; c1++) {
311 0 : histp = & histogram[c0][c1][c2min];
312 0 : for (c2 = c2min; c2 <= c2max; c2++)
313 0 : if (*histp++ != 0) {
314 0 : boxp->c0min = c0min = c0;
315 0 : goto have_c0min;
316 : }
317 : }
318 : have_c0min:
319 0 : if (c0max > c0min)
320 0 : for (c0 = c0max; c0 >= c0min; c0--)
321 0 : for (c1 = c1min; c1 <= c1max; c1++) {
322 0 : histp = & histogram[c0][c1][c2min];
323 0 : for (c2 = c2min; c2 <= c2max; c2++)
324 0 : if (*histp++ != 0) {
325 0 : boxp->c0max = c0max = c0;
326 0 : goto have_c0max;
327 : }
328 : }
329 : have_c0max:
330 0 : if (c1max > c1min)
331 0 : for (c1 = c1min; c1 <= c1max; c1++)
332 0 : for (c0 = c0min; c0 <= c0max; c0++) {
333 0 : histp = & histogram[c0][c1][c2min];
334 0 : for (c2 = c2min; c2 <= c2max; c2++)
335 0 : if (*histp++ != 0) {
336 0 : boxp->c1min = c1min = c1;
337 0 : goto have_c1min;
338 : }
339 : }
340 : have_c1min:
341 0 : if (c1max > c1min)
342 0 : for (c1 = c1max; c1 >= c1min; c1--)
343 0 : for (c0 = c0min; c0 <= c0max; c0++) {
344 0 : histp = & histogram[c0][c1][c2min];
345 0 : for (c2 = c2min; c2 <= c2max; c2++)
346 0 : if (*histp++ != 0) {
347 0 : boxp->c1max = c1max = c1;
348 0 : goto have_c1max;
349 : }
350 : }
351 : have_c1max:
352 0 : if (c2max > c2min)
353 0 : for (c2 = c2min; c2 <= c2max; c2++)
354 0 : for (c0 = c0min; c0 <= c0max; c0++) {
355 0 : histp = & histogram[c0][c1min][c2];
356 0 : for (c1 = c1min; c1 <= c1max; c1++, histp += HIST_C2_ELEMS)
357 0 : if (*histp != 0) {
358 0 : boxp->c2min = c2min = c2;
359 0 : goto have_c2min;
360 : }
361 : }
362 : have_c2min:
363 0 : if (c2max > c2min)
364 0 : for (c2 = c2max; c2 >= c2min; c2--)
365 0 : for (c0 = c0min; c0 <= c0max; c0++) {
366 0 : histp = & histogram[c0][c1min][c2];
367 0 : for (c1 = c1min; c1 <= c1max; c1++, histp += HIST_C2_ELEMS)
368 0 : if (*histp != 0) {
369 0 : boxp->c2max = c2max = c2;
370 0 : goto have_c2max;
371 : }
372 : }
373 : have_c2max:
374 :
375 : /* Update box volume.
376 : * We use 2-norm rather than real volume here; this biases the method
377 : * against making long narrow boxes, and it has the side benefit that
378 : * a box is splittable iff norm > 0.
379 : * Since the differences are expressed in histogram-cell units,
380 : * we have to shift back to JSAMPLE units to get consistent distances;
381 : * after which, we scale according to the selected distance scale factors.
382 : */
383 0 : dist0 = ((c0max - c0min) << C0_SHIFT) * C0_SCALE;
384 0 : dist1 = ((c1max - c1min) << C1_SHIFT) * C1_SCALE;
385 0 : dist2 = ((c2max - c2min) << C2_SHIFT) * C2_SCALE;
386 0 : boxp->volume = dist0*dist0 + dist1*dist1 + dist2*dist2;
387 :
388 : /* Now scan remaining volume of box and compute population */
389 0 : ccount = 0;
390 0 : for (c0 = c0min; c0 <= c0max; c0++)
391 0 : for (c1 = c1min; c1 <= c1max; c1++) {
392 0 : histp = & histogram[c0][c1][c2min];
393 0 : for (c2 = c2min; c2 <= c2max; c2++, histp++)
394 0 : if (*histp != 0) {
395 0 : ccount++;
396 : }
397 : }
398 0 : boxp->colorcount = ccount;
399 0 : }
400 :
401 :
402 : LOCAL(int)
403 0 : median_cut (j_decompress_ptr cinfo, boxptr boxlist, int numboxes,
404 : int desired_colors)
405 : /* Repeatedly select and split the largest box until we have enough boxes */
406 : {
407 : int n,lb;
408 : int c0,c1,c2,cmax;
409 : register boxptr b1,b2;
410 :
411 0 : while (numboxes < desired_colors) {
412 : /* Select box to split.
413 : * Current algorithm: by population for first half, then by volume.
414 : */
415 0 : if (numboxes*2 <= desired_colors) {
416 0 : b1 = find_biggest_color_pop(boxlist, numboxes);
417 : } else {
418 0 : b1 = find_biggest_volume(boxlist, numboxes);
419 : }
420 0 : if (b1 == NULL) /* no splittable boxes left! */
421 0 : break;
422 0 : b2 = &boxlist[numboxes]; /* where new box will go */
423 : /* Copy the color bounds to the new box. */
424 0 : b2->c0max = b1->c0max; b2->c1max = b1->c1max; b2->c2max = b1->c2max;
425 0 : b2->c0min = b1->c0min; b2->c1min = b1->c1min; b2->c2min = b1->c2min;
426 : /* Choose which axis to split the box on.
427 : * Current algorithm: longest scaled axis.
428 : * See notes in update_box about scaling distances.
429 : */
430 0 : c0 = ((b1->c0max - b1->c0min) << C0_SHIFT) * C0_SCALE;
431 0 : c1 = ((b1->c1max - b1->c1min) << C1_SHIFT) * C1_SCALE;
432 0 : c2 = ((b1->c2max - b1->c2min) << C2_SHIFT) * C2_SCALE;
433 : /* We want to break any ties in favor of green, then red, blue last.
434 : * This code does the right thing for R,G,B or B,G,R color orders only.
435 : */
436 0 : if (rgb_red[cinfo->out_color_space] == 0) {
437 0 : cmax = c1; n = 1;
438 0 : if (c0 > cmax) { cmax = c0; n = 0; }
439 0 : if (c2 > cmax) { n = 2; }
440 : }
441 : else {
442 0 : cmax = c1; n = 1;
443 0 : if (c2 > cmax) { cmax = c2; n = 2; }
444 0 : if (c0 > cmax) { n = 0; }
445 : }
446 : /* Choose split point along selected axis, and update box bounds.
447 : * Current algorithm: split at halfway point.
448 : * (Since the box has been shrunk to minimum volume,
449 : * any split will produce two nonempty subboxes.)
450 : * Note that lb value is max for lower box, so must be < old max.
451 : */
452 0 : switch (n) {
453 : case 0:
454 0 : lb = (b1->c0max + b1->c0min) / 2;
455 0 : b1->c0max = lb;
456 0 : b2->c0min = lb+1;
457 0 : break;
458 : case 1:
459 0 : lb = (b1->c1max + b1->c1min) / 2;
460 0 : b1->c1max = lb;
461 0 : b2->c1min = lb+1;
462 0 : break;
463 : case 2:
464 0 : lb = (b1->c2max + b1->c2min) / 2;
465 0 : b1->c2max = lb;
466 0 : b2->c2min = lb+1;
467 0 : break;
468 : }
469 : /* Update stats for boxes */
470 0 : update_box(cinfo, b1);
471 0 : update_box(cinfo, b2);
472 0 : numboxes++;
473 : }
474 0 : return numboxes;
475 : }
476 :
477 :
478 : LOCAL(void)
479 0 : compute_color (j_decompress_ptr cinfo, boxptr boxp, int icolor)
480 : /* Compute representative color for a box, put it in colormap[icolor] */
481 : {
482 : /* Current algorithm: mean weighted by pixels (not colors) */
483 : /* Note it is important to get the rounding correct! */
484 0 : my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
485 0 : hist3d histogram = cquantize->histogram;
486 : histptr histp;
487 : int c0,c1,c2;
488 : int c0min,c0max,c1min,c1max,c2min,c2max;
489 : long count;
490 0 : long total = 0;
491 0 : long c0total = 0;
492 0 : long c1total = 0;
493 0 : long c2total = 0;
494 :
495 0 : c0min = boxp->c0min; c0max = boxp->c0max;
496 0 : c1min = boxp->c1min; c1max = boxp->c1max;
497 0 : c2min = boxp->c2min; c2max = boxp->c2max;
498 :
499 0 : for (c0 = c0min; c0 <= c0max; c0++)
500 0 : for (c1 = c1min; c1 <= c1max; c1++) {
501 0 : histp = & histogram[c0][c1][c2min];
502 0 : for (c2 = c2min; c2 <= c2max; c2++) {
503 0 : if ((count = *histp++) != 0) {
504 0 : total += count;
505 0 : c0total += ((c0 << C0_SHIFT) + ((1<<C0_SHIFT)>>1)) * count;
506 0 : c1total += ((c1 << C1_SHIFT) + ((1<<C1_SHIFT)>>1)) * count;
507 0 : c2total += ((c2 << C2_SHIFT) + ((1<<C2_SHIFT)>>1)) * count;
508 : }
509 : }
510 : }
511 :
512 0 : cinfo->colormap[0][icolor] = (JSAMPLE) ((c0total + (total>>1)) / total);
513 0 : cinfo->colormap[1][icolor] = (JSAMPLE) ((c1total + (total>>1)) / total);
514 0 : cinfo->colormap[2][icolor] = (JSAMPLE) ((c2total + (total>>1)) / total);
515 0 : }
516 :
517 :
518 : LOCAL(void)
519 0 : select_colors (j_decompress_ptr cinfo, int desired_colors)
520 : /* Master routine for color selection */
521 : {
522 : boxptr boxlist;
523 : int numboxes;
524 : int i;
525 :
526 : /* Allocate workspace for box list */
527 0 : boxlist = (boxptr) (*cinfo->mem->alloc_small)
528 : ((j_common_ptr) cinfo, JPOOL_IMAGE, desired_colors * sizeof(box));
529 : /* Initialize one box containing whole space */
530 0 : numboxes = 1;
531 0 : boxlist[0].c0min = 0;
532 0 : boxlist[0].c0max = MAXJSAMPLE >> C0_SHIFT;
533 0 : boxlist[0].c1min = 0;
534 0 : boxlist[0].c1max = MAXJSAMPLE >> C1_SHIFT;
535 0 : boxlist[0].c2min = 0;
536 0 : boxlist[0].c2max = MAXJSAMPLE >> C2_SHIFT;
537 : /* Shrink it to actually-used volume and set its statistics */
538 0 : update_box(cinfo, & boxlist[0]);
539 : /* Perform median-cut to produce final box list */
540 0 : numboxes = median_cut(cinfo, boxlist, numboxes, desired_colors);
541 : /* Compute the representative color for each box, fill colormap */
542 0 : for (i = 0; i < numboxes; i++)
543 0 : compute_color(cinfo, & boxlist[i], i);
544 0 : cinfo->actual_number_of_colors = numboxes;
545 0 : TRACEMS1(cinfo, 1, JTRC_QUANT_SELECTED, numboxes);
546 0 : }
547 :
548 :
549 : /*
550 : * These routines are concerned with the time-critical task of mapping input
551 : * colors to the nearest color in the selected colormap.
552 : *
553 : * We re-use the histogram space as an "inverse color map", essentially a
554 : * cache for the results of nearest-color searches. All colors within a
555 : * histogram cell will be mapped to the same colormap entry, namely the one
556 : * closest to the cell's center. This may not be quite the closest entry to
557 : * the actual input color, but it's almost as good. A zero in the cache
558 : * indicates we haven't found the nearest color for that cell yet; the array
559 : * is cleared to zeroes before starting the mapping pass. When we find the
560 : * nearest color for a cell, its colormap index plus one is recorded in the
561 : * cache for future use. The pass2 scanning routines call fill_inverse_cmap
562 : * when they need to use an unfilled entry in the cache.
563 : *
564 : * Our method of efficiently finding nearest colors is based on the "locally
565 : * sorted search" idea described by Heckbert and on the incremental distance
566 : * calculation described by Spencer W. Thomas in chapter III.1 of Graphics
567 : * Gems II (James Arvo, ed. Academic Press, 1991). Thomas points out that
568 : * the distances from a given colormap entry to each cell of the histogram can
569 : * be computed quickly using an incremental method: the differences between
570 : * distances to adjacent cells themselves differ by a constant. This allows a
571 : * fairly fast implementation of the "brute force" approach of computing the
572 : * distance from every colormap entry to every histogram cell. Unfortunately,
573 : * it needs a work array to hold the best-distance-so-far for each histogram
574 : * cell (because the inner loop has to be over cells, not colormap entries).
575 : * The work array elements have to be JLONGs, so the work array would need
576 : * 256Kb at our recommended precision. This is not feasible in DOS machines.
577 : *
578 : * To get around these problems, we apply Thomas' method to compute the
579 : * nearest colors for only the cells within a small subbox of the histogram.
580 : * The work array need be only as big as the subbox, so the memory usage
581 : * problem is solved. Furthermore, we need not fill subboxes that are never
582 : * referenced in pass2; many images use only part of the color gamut, so a
583 : * fair amount of work is saved. An additional advantage of this
584 : * approach is that we can apply Heckbert's locality criterion to quickly
585 : * eliminate colormap entries that are far away from the subbox; typically
586 : * three-fourths of the colormap entries are rejected by Heckbert's criterion,
587 : * and we need not compute their distances to individual cells in the subbox.
588 : * The speed of this approach is heavily influenced by the subbox size: too
589 : * small means too much overhead, too big loses because Heckbert's criterion
590 : * can't eliminate as many colormap entries. Empirically the best subbox
591 : * size seems to be about 1/512th of the histogram (1/8th in each direction).
592 : *
593 : * Thomas' article also describes a refined method which is asymptotically
594 : * faster than the brute-force method, but it is also far more complex and
595 : * cannot efficiently be applied to small subboxes. It is therefore not
596 : * useful for programs intended to be portable to DOS machines. On machines
597 : * with plenty of memory, filling the whole histogram in one shot with Thomas'
598 : * refined method might be faster than the present code --- but then again,
599 : * it might not be any faster, and it's certainly more complicated.
600 : */
601 :
602 :
603 : /* log2(histogram cells in update box) for each axis; this can be adjusted */
604 : #define BOX_C0_LOG (HIST_C0_BITS-3)
605 : #define BOX_C1_LOG (HIST_C1_BITS-3)
606 : #define BOX_C2_LOG (HIST_C2_BITS-3)
607 :
608 : #define BOX_C0_ELEMS (1<<BOX_C0_LOG) /* # of hist cells in update box */
609 : #define BOX_C1_ELEMS (1<<BOX_C1_LOG)
610 : #define BOX_C2_ELEMS (1<<BOX_C2_LOG)
611 :
612 : #define BOX_C0_SHIFT (C0_SHIFT + BOX_C0_LOG)
613 : #define BOX_C1_SHIFT (C1_SHIFT + BOX_C1_LOG)
614 : #define BOX_C2_SHIFT (C2_SHIFT + BOX_C2_LOG)
615 :
616 :
617 : /*
618 : * The next three routines implement inverse colormap filling. They could
619 : * all be folded into one big routine, but splitting them up this way saves
620 : * some stack space (the mindist[] and bestdist[] arrays need not coexist)
621 : * and may allow some compilers to produce better code by registerizing more
622 : * inner-loop variables.
623 : */
624 :
625 : LOCAL(int)
626 0 : find_nearby_colors (j_decompress_ptr cinfo, int minc0, int minc1, int minc2,
627 : JSAMPLE colorlist[])
628 : /* Locate the colormap entries close enough to an update box to be candidates
629 : * for the nearest entry to some cell(s) in the update box. The update box
630 : * is specified by the center coordinates of its first cell. The number of
631 : * candidate colormap entries is returned, and their colormap indexes are
632 : * placed in colorlist[].
633 : * This routine uses Heckbert's "locally sorted search" criterion to select
634 : * the colors that need further consideration.
635 : */
636 : {
637 0 : int numcolors = cinfo->actual_number_of_colors;
638 : int maxc0, maxc1, maxc2;
639 : int centerc0, centerc1, centerc2;
640 : int i, x, ncolors;
641 : JLONG minmaxdist, min_dist, max_dist, tdist;
642 : JLONG mindist[MAXNUMCOLORS]; /* min distance to colormap entry i */
643 :
644 : /* Compute true coordinates of update box's upper corner and center.
645 : * Actually we compute the coordinates of the center of the upper-corner
646 : * histogram cell, which are the upper bounds of the volume we care about.
647 : * Note that since ">>" rounds down, the "center" values may be closer to
648 : * min than to max; hence comparisons to them must be "<=", not "<".
649 : */
650 0 : maxc0 = minc0 + ((1 << BOX_C0_SHIFT) - (1 << C0_SHIFT));
651 0 : centerc0 = (minc0 + maxc0) >> 1;
652 0 : maxc1 = minc1 + ((1 << BOX_C1_SHIFT) - (1 << C1_SHIFT));
653 0 : centerc1 = (minc1 + maxc1) >> 1;
654 0 : maxc2 = minc2 + ((1 << BOX_C2_SHIFT) - (1 << C2_SHIFT));
655 0 : centerc2 = (minc2 + maxc2) >> 1;
656 :
657 : /* For each color in colormap, find:
658 : * 1. its minimum squared-distance to any point in the update box
659 : * (zero if color is within update box);
660 : * 2. its maximum squared-distance to any point in the update box.
661 : * Both of these can be found by considering only the corners of the box.
662 : * We save the minimum distance for each color in mindist[];
663 : * only the smallest maximum distance is of interest.
664 : */
665 0 : minmaxdist = 0x7FFFFFFFL;
666 :
667 0 : for (i = 0; i < numcolors; i++) {
668 : /* We compute the squared-c0-distance term, then add in the other two. */
669 0 : x = GETJSAMPLE(cinfo->colormap[0][i]);
670 0 : if (x < minc0) {
671 0 : tdist = (x - minc0) * C0_SCALE;
672 0 : min_dist = tdist*tdist;
673 0 : tdist = (x - maxc0) * C0_SCALE;
674 0 : max_dist = tdist*tdist;
675 0 : } else if (x > maxc0) {
676 0 : tdist = (x - maxc0) * C0_SCALE;
677 0 : min_dist = tdist*tdist;
678 0 : tdist = (x - minc0) * C0_SCALE;
679 0 : max_dist = tdist*tdist;
680 : } else {
681 : /* within cell range so no contribution to min_dist */
682 0 : min_dist = 0;
683 0 : if (x <= centerc0) {
684 0 : tdist = (x - maxc0) * C0_SCALE;
685 0 : max_dist = tdist*tdist;
686 : } else {
687 0 : tdist = (x - minc0) * C0_SCALE;
688 0 : max_dist = tdist*tdist;
689 : }
690 : }
691 :
692 0 : x = GETJSAMPLE(cinfo->colormap[1][i]);
693 0 : if (x < minc1) {
694 0 : tdist = (x - minc1) * C1_SCALE;
695 0 : min_dist += tdist*tdist;
696 0 : tdist = (x - maxc1) * C1_SCALE;
697 0 : max_dist += tdist*tdist;
698 0 : } else if (x > maxc1) {
699 0 : tdist = (x - maxc1) * C1_SCALE;
700 0 : min_dist += tdist*tdist;
701 0 : tdist = (x - minc1) * C1_SCALE;
702 0 : max_dist += tdist*tdist;
703 : } else {
704 : /* within cell range so no contribution to min_dist */
705 0 : if (x <= centerc1) {
706 0 : tdist = (x - maxc1) * C1_SCALE;
707 0 : max_dist += tdist*tdist;
708 : } else {
709 0 : tdist = (x - minc1) * C1_SCALE;
710 0 : max_dist += tdist*tdist;
711 : }
712 : }
713 :
714 0 : x = GETJSAMPLE(cinfo->colormap[2][i]);
715 0 : if (x < minc2) {
716 0 : tdist = (x - minc2) * C2_SCALE;
717 0 : min_dist += tdist*tdist;
718 0 : tdist = (x - maxc2) * C2_SCALE;
719 0 : max_dist += tdist*tdist;
720 0 : } else if (x > maxc2) {
721 0 : tdist = (x - maxc2) * C2_SCALE;
722 0 : min_dist += tdist*tdist;
723 0 : tdist = (x - minc2) * C2_SCALE;
724 0 : max_dist += tdist*tdist;
725 : } else {
726 : /* within cell range so no contribution to min_dist */
727 0 : if (x <= centerc2) {
728 0 : tdist = (x - maxc2) * C2_SCALE;
729 0 : max_dist += tdist*tdist;
730 : } else {
731 0 : tdist = (x - minc2) * C2_SCALE;
732 0 : max_dist += tdist*tdist;
733 : }
734 : }
735 :
736 0 : mindist[i] = min_dist; /* save away the results */
737 0 : if (max_dist < minmaxdist)
738 0 : minmaxdist = max_dist;
739 : }
740 :
741 : /* Now we know that no cell in the update box is more than minmaxdist
742 : * away from some colormap entry. Therefore, only colors that are
743 : * within minmaxdist of some part of the box need be considered.
744 : */
745 0 : ncolors = 0;
746 0 : for (i = 0; i < numcolors; i++) {
747 0 : if (mindist[i] <= minmaxdist)
748 0 : colorlist[ncolors++] = (JSAMPLE) i;
749 : }
750 0 : return ncolors;
751 : }
752 :
753 :
754 : LOCAL(void)
755 0 : find_best_colors (j_decompress_ptr cinfo, int minc0, int minc1, int minc2,
756 : int numcolors, JSAMPLE colorlist[], JSAMPLE bestcolor[])
757 : /* Find the closest colormap entry for each cell in the update box,
758 : * given the list of candidate colors prepared by find_nearby_colors.
759 : * Return the indexes of the closest entries in the bestcolor[] array.
760 : * This routine uses Thomas' incremental distance calculation method to
761 : * find the distance from a colormap entry to successive cells in the box.
762 : */
763 : {
764 : int ic0, ic1, ic2;
765 : int i, icolor;
766 : register JLONG *bptr; /* pointer into bestdist[] array */
767 : JSAMPLE *cptr; /* pointer into bestcolor[] array */
768 : JLONG dist0, dist1; /* initial distance values */
769 : register JLONG dist2; /* current distance in inner loop */
770 : JLONG xx0, xx1; /* distance increments */
771 : register JLONG xx2;
772 : JLONG inc0, inc1, inc2; /* initial values for increments */
773 : /* This array holds the distance to the nearest-so-far color for each cell */
774 : JLONG bestdist[BOX_C0_ELEMS * BOX_C1_ELEMS * BOX_C2_ELEMS];
775 :
776 : /* Initialize best-distance for each cell of the update box */
777 0 : bptr = bestdist;
778 0 : for (i = BOX_C0_ELEMS*BOX_C1_ELEMS*BOX_C2_ELEMS-1; i >= 0; i--)
779 0 : *bptr++ = 0x7FFFFFFFL;
780 :
781 : /* For each color selected by find_nearby_colors,
782 : * compute its distance to the center of each cell in the box.
783 : * If that's less than best-so-far, update best distance and color number.
784 : */
785 :
786 : /* Nominal steps between cell centers ("x" in Thomas article) */
787 : #define STEP_C0 ((1 << C0_SHIFT) * C0_SCALE)
788 : #define STEP_C1 ((1 << C1_SHIFT) * C1_SCALE)
789 : #define STEP_C2 ((1 << C2_SHIFT) * C2_SCALE)
790 :
791 0 : for (i = 0; i < numcolors; i++) {
792 0 : icolor = GETJSAMPLE(colorlist[i]);
793 : /* Compute (square of) distance from minc0/c1/c2 to this color */
794 0 : inc0 = (minc0 - GETJSAMPLE(cinfo->colormap[0][icolor])) * C0_SCALE;
795 0 : dist0 = inc0*inc0;
796 0 : inc1 = (minc1 - GETJSAMPLE(cinfo->colormap[1][icolor])) * C1_SCALE;
797 0 : dist0 += inc1*inc1;
798 0 : inc2 = (minc2 - GETJSAMPLE(cinfo->colormap[2][icolor])) * C2_SCALE;
799 0 : dist0 += inc2*inc2;
800 : /* Form the initial difference increments */
801 0 : inc0 = inc0 * (2 * STEP_C0) + STEP_C0 * STEP_C0;
802 0 : inc1 = inc1 * (2 * STEP_C1) + STEP_C1 * STEP_C1;
803 0 : inc2 = inc2 * (2 * STEP_C2) + STEP_C2 * STEP_C2;
804 : /* Now loop over all cells in box, updating distance per Thomas method */
805 0 : bptr = bestdist;
806 0 : cptr = bestcolor;
807 0 : xx0 = inc0;
808 0 : for (ic0 = BOX_C0_ELEMS-1; ic0 >= 0; ic0--) {
809 0 : dist1 = dist0;
810 0 : xx1 = inc1;
811 0 : for (ic1 = BOX_C1_ELEMS-1; ic1 >= 0; ic1--) {
812 0 : dist2 = dist1;
813 0 : xx2 = inc2;
814 0 : for (ic2 = BOX_C2_ELEMS-1; ic2 >= 0; ic2--) {
815 0 : if (dist2 < *bptr) {
816 0 : *bptr = dist2;
817 0 : *cptr = (JSAMPLE) icolor;
818 : }
819 0 : dist2 += xx2;
820 0 : xx2 += 2 * STEP_C2 * STEP_C2;
821 0 : bptr++;
822 0 : cptr++;
823 : }
824 0 : dist1 += xx1;
825 0 : xx1 += 2 * STEP_C1 * STEP_C1;
826 : }
827 0 : dist0 += xx0;
828 0 : xx0 += 2 * STEP_C0 * STEP_C0;
829 : }
830 : }
831 0 : }
832 :
833 :
834 : LOCAL(void)
835 0 : fill_inverse_cmap (j_decompress_ptr cinfo, int c0, int c1, int c2)
836 : /* Fill the inverse-colormap entries in the update box that contains */
837 : /* histogram cell c0/c1/c2. (Only that one cell MUST be filled, but */
838 : /* we can fill as many others as we wish.) */
839 : {
840 0 : my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
841 0 : hist3d histogram = cquantize->histogram;
842 : int minc0, minc1, minc2; /* lower left corner of update box */
843 : int ic0, ic1, ic2;
844 : register JSAMPLE *cptr; /* pointer into bestcolor[] array */
845 : register histptr cachep; /* pointer into main cache array */
846 : /* This array lists the candidate colormap indexes. */
847 : JSAMPLE colorlist[MAXNUMCOLORS];
848 : int numcolors; /* number of candidate colors */
849 : /* This array holds the actually closest colormap index for each cell. */
850 : JSAMPLE bestcolor[BOX_C0_ELEMS * BOX_C1_ELEMS * BOX_C2_ELEMS];
851 :
852 : /* Convert cell coordinates to update box ID */
853 0 : c0 >>= BOX_C0_LOG;
854 0 : c1 >>= BOX_C1_LOG;
855 0 : c2 >>= BOX_C2_LOG;
856 :
857 : /* Compute true coordinates of update box's origin corner.
858 : * Actually we compute the coordinates of the center of the corner
859 : * histogram cell, which are the lower bounds of the volume we care about.
860 : */
861 0 : minc0 = (c0 << BOX_C0_SHIFT) + ((1 << C0_SHIFT) >> 1);
862 0 : minc1 = (c1 << BOX_C1_SHIFT) + ((1 << C1_SHIFT) >> 1);
863 0 : minc2 = (c2 << BOX_C2_SHIFT) + ((1 << C2_SHIFT) >> 1);
864 :
865 : /* Determine which colormap entries are close enough to be candidates
866 : * for the nearest entry to some cell in the update box.
867 : */
868 0 : numcolors = find_nearby_colors(cinfo, minc0, minc1, minc2, colorlist);
869 :
870 : /* Determine the actually nearest colors. */
871 0 : find_best_colors(cinfo, minc0, minc1, minc2, numcolors, colorlist,
872 : bestcolor);
873 :
874 : /* Save the best color numbers (plus 1) in the main cache array */
875 0 : c0 <<= BOX_C0_LOG; /* convert ID back to base cell indexes */
876 0 : c1 <<= BOX_C1_LOG;
877 0 : c2 <<= BOX_C2_LOG;
878 0 : cptr = bestcolor;
879 0 : for (ic0 = 0; ic0 < BOX_C0_ELEMS; ic0++) {
880 0 : for (ic1 = 0; ic1 < BOX_C1_ELEMS; ic1++) {
881 0 : cachep = & histogram[c0+ic0][c1+ic1][c2];
882 0 : for (ic2 = 0; ic2 < BOX_C2_ELEMS; ic2++) {
883 0 : *cachep++ = (histcell) (GETJSAMPLE(*cptr++) + 1);
884 : }
885 : }
886 : }
887 0 : }
888 :
889 :
890 : /*
891 : * Map some rows of pixels to the output colormapped representation.
892 : */
893 :
894 : METHODDEF(void)
895 0 : pass2_no_dither (j_decompress_ptr cinfo,
896 : JSAMPARRAY input_buf, JSAMPARRAY output_buf, int num_rows)
897 : /* This version performs no dithering */
898 : {
899 0 : my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
900 0 : hist3d histogram = cquantize->histogram;
901 : register JSAMPROW inptr, outptr;
902 : register histptr cachep;
903 : register int c0, c1, c2;
904 : int row;
905 : JDIMENSION col;
906 0 : JDIMENSION width = cinfo->output_width;
907 :
908 0 : for (row = 0; row < num_rows; row++) {
909 0 : inptr = input_buf[row];
910 0 : outptr = output_buf[row];
911 0 : for (col = width; col > 0; col--) {
912 : /* get pixel value and index into the cache */
913 0 : c0 = GETJSAMPLE(*inptr++) >> C0_SHIFT;
914 0 : c1 = GETJSAMPLE(*inptr++) >> C1_SHIFT;
915 0 : c2 = GETJSAMPLE(*inptr++) >> C2_SHIFT;
916 0 : cachep = & histogram[c0][c1][c2];
917 : /* If we have not seen this color before, find nearest colormap entry */
918 : /* and update the cache */
919 0 : if (*cachep == 0)
920 0 : fill_inverse_cmap(cinfo, c0,c1,c2);
921 : /* Now emit the colormap index for this cell */
922 0 : *outptr++ = (JSAMPLE) (*cachep - 1);
923 : }
924 : }
925 0 : }
926 :
927 :
928 : METHODDEF(void)
929 0 : pass2_fs_dither (j_decompress_ptr cinfo,
930 : JSAMPARRAY input_buf, JSAMPARRAY output_buf, int num_rows)
931 : /* This version performs Floyd-Steinberg dithering */
932 : {
933 0 : my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
934 0 : hist3d histogram = cquantize->histogram;
935 : register LOCFSERROR cur0, cur1, cur2; /* current error or pixel value */
936 : LOCFSERROR belowerr0, belowerr1, belowerr2; /* error for pixel below cur */
937 : LOCFSERROR bpreverr0, bpreverr1, bpreverr2; /* error for below/prev col */
938 : register FSERRPTR errorptr; /* => fserrors[] at column before current */
939 : JSAMPROW inptr; /* => current input pixel */
940 : JSAMPROW outptr; /* => current output pixel */
941 : histptr cachep;
942 : int dir; /* +1 or -1 depending on direction */
943 : int dir3; /* 3*dir, for advancing inptr & errorptr */
944 : int row;
945 : JDIMENSION col;
946 0 : JDIMENSION width = cinfo->output_width;
947 0 : JSAMPLE *range_limit = cinfo->sample_range_limit;
948 0 : int *error_limit = cquantize->error_limiter;
949 0 : JSAMPROW colormap0 = cinfo->colormap[0];
950 0 : JSAMPROW colormap1 = cinfo->colormap[1];
951 0 : JSAMPROW colormap2 = cinfo->colormap[2];
952 : SHIFT_TEMPS
953 :
954 0 : for (row = 0; row < num_rows; row++) {
955 0 : inptr = input_buf[row];
956 0 : outptr = output_buf[row];
957 0 : if (cquantize->on_odd_row) {
958 : /* work right to left in this row */
959 0 : inptr += (width-1) * 3; /* so point to rightmost pixel */
960 0 : outptr += width-1;
961 0 : dir = -1;
962 0 : dir3 = -3;
963 0 : errorptr = cquantize->fserrors + (width+1)*3; /* => entry after last column */
964 0 : cquantize->on_odd_row = FALSE; /* flip for next time */
965 : } else {
966 : /* work left to right in this row */
967 0 : dir = 1;
968 0 : dir3 = 3;
969 0 : errorptr = cquantize->fserrors; /* => entry before first real column */
970 0 : cquantize->on_odd_row = TRUE; /* flip for next time */
971 : }
972 : /* Preset error values: no error propagated to first pixel from left */
973 0 : cur0 = cur1 = cur2 = 0;
974 : /* and no error propagated to row below yet */
975 0 : belowerr0 = belowerr1 = belowerr2 = 0;
976 0 : bpreverr0 = bpreverr1 = bpreverr2 = 0;
977 :
978 0 : for (col = width; col > 0; col--) {
979 : /* curN holds the error propagated from the previous pixel on the
980 : * current line. Add the error propagated from the previous line
981 : * to form the complete error correction term for this pixel, and
982 : * round the error term (which is expressed * 16) to an integer.
983 : * RIGHT_SHIFT rounds towards minus infinity, so adding 8 is correct
984 : * for either sign of the error value.
985 : * Note: errorptr points to *previous* column's array entry.
986 : */
987 0 : cur0 = RIGHT_SHIFT(cur0 + errorptr[dir3+0] + 8, 4);
988 0 : cur1 = RIGHT_SHIFT(cur1 + errorptr[dir3+1] + 8, 4);
989 0 : cur2 = RIGHT_SHIFT(cur2 + errorptr[dir3+2] + 8, 4);
990 : /* Limit the error using transfer function set by init_error_limit.
991 : * See comments with init_error_limit for rationale.
992 : */
993 0 : cur0 = error_limit[cur0];
994 0 : cur1 = error_limit[cur1];
995 0 : cur2 = error_limit[cur2];
996 : /* Form pixel value + error, and range-limit to 0..MAXJSAMPLE.
997 : * The maximum error is +- MAXJSAMPLE (or less with error limiting);
998 : * this sets the required size of the range_limit array.
999 : */
1000 0 : cur0 += GETJSAMPLE(inptr[0]);
1001 0 : cur1 += GETJSAMPLE(inptr[1]);
1002 0 : cur2 += GETJSAMPLE(inptr[2]);
1003 0 : cur0 = GETJSAMPLE(range_limit[cur0]);
1004 0 : cur1 = GETJSAMPLE(range_limit[cur1]);
1005 0 : cur2 = GETJSAMPLE(range_limit[cur2]);
1006 : /* Index into the cache with adjusted pixel value */
1007 0 : cachep = & histogram[cur0>>C0_SHIFT][cur1>>C1_SHIFT][cur2>>C2_SHIFT];
1008 : /* If we have not seen this color before, find nearest colormap */
1009 : /* entry and update the cache */
1010 0 : if (*cachep == 0)
1011 0 : fill_inverse_cmap(cinfo, cur0>>C0_SHIFT,cur1>>C1_SHIFT,cur2>>C2_SHIFT);
1012 : /* Now emit the colormap index for this cell */
1013 0 : { register int pixcode = *cachep - 1;
1014 0 : *outptr = (JSAMPLE) pixcode;
1015 : /* Compute representation error for this pixel */
1016 0 : cur0 -= GETJSAMPLE(colormap0[pixcode]);
1017 0 : cur1 -= GETJSAMPLE(colormap1[pixcode]);
1018 0 : cur2 -= GETJSAMPLE(colormap2[pixcode]);
1019 : }
1020 : /* Compute error fractions to be propagated to adjacent pixels.
1021 : * Add these into the running sums, and simultaneously shift the
1022 : * next-line error sums left by 1 column.
1023 : */
1024 : { register LOCFSERROR bnexterr;
1025 :
1026 0 : bnexterr = cur0; /* Process component 0 */
1027 0 : errorptr[0] = (FSERROR) (bpreverr0 + cur0 * 3);
1028 0 : bpreverr0 = belowerr0 + cur0 * 5;
1029 0 : belowerr0 = bnexterr;
1030 0 : cur0 *= 7;
1031 0 : bnexterr = cur1; /* Process component 1 */
1032 0 : errorptr[1] = (FSERROR) (bpreverr1 + cur1 * 3);
1033 0 : bpreverr1 = belowerr1 + cur1 * 5;
1034 0 : belowerr1 = bnexterr;
1035 0 : cur1 *= 7;
1036 0 : bnexterr = cur2; /* Process component 2 */
1037 0 : errorptr[2] = (FSERROR) (bpreverr2 + cur2 * 3);
1038 0 : bpreverr2 = belowerr2 + cur2 * 5;
1039 0 : belowerr2 = bnexterr;
1040 0 : cur2 *= 7;
1041 : }
1042 : /* At this point curN contains the 7/16 error value to be propagated
1043 : * to the next pixel on the current line, and all the errors for the
1044 : * next line have been shifted over. We are therefore ready to move on.
1045 : */
1046 0 : inptr += dir3; /* Advance pixel pointers to next column */
1047 0 : outptr += dir;
1048 0 : errorptr += dir3; /* advance errorptr to current column */
1049 : }
1050 : /* Post-loop cleanup: we must unload the final error values into the
1051 : * final fserrors[] entry. Note we need not unload belowerrN because
1052 : * it is for the dummy column before or after the actual array.
1053 : */
1054 0 : errorptr[0] = (FSERROR) bpreverr0; /* unload prev errs into array */
1055 0 : errorptr[1] = (FSERROR) bpreverr1;
1056 0 : errorptr[2] = (FSERROR) bpreverr2;
1057 : }
1058 0 : }
1059 :
1060 :
1061 : /*
1062 : * Initialize the error-limiting transfer function (lookup table).
1063 : * The raw F-S error computation can potentially compute error values of up to
1064 : * +- MAXJSAMPLE. But we want the maximum correction applied to a pixel to be
1065 : * much less, otherwise obviously wrong pixels will be created. (Typical
1066 : * effects include weird fringes at color-area boundaries, isolated bright
1067 : * pixels in a dark area, etc.) The standard advice for avoiding this problem
1068 : * is to ensure that the "corners" of the color cube are allocated as output
1069 : * colors; then repeated errors in the same direction cannot cause cascading
1070 : * error buildup. However, that only prevents the error from getting
1071 : * completely out of hand; Aaron Giles reports that error limiting improves
1072 : * the results even with corner colors allocated.
1073 : * A simple clamping of the error values to about +- MAXJSAMPLE/8 works pretty
1074 : * well, but the smoother transfer function used below is even better. Thanks
1075 : * to Aaron Giles for this idea.
1076 : */
1077 :
1078 : LOCAL(void)
1079 0 : init_error_limit (j_decompress_ptr cinfo)
1080 : /* Allocate and fill in the error_limiter table */
1081 : {
1082 0 : my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
1083 : int *table;
1084 : int in, out;
1085 :
1086 0 : table = (int *) (*cinfo->mem->alloc_small)
1087 : ((j_common_ptr) cinfo, JPOOL_IMAGE, (MAXJSAMPLE*2+1) * sizeof(int));
1088 0 : table += MAXJSAMPLE; /* so can index -MAXJSAMPLE .. +MAXJSAMPLE */
1089 0 : cquantize->error_limiter = table;
1090 :
1091 : #define STEPSIZE ((MAXJSAMPLE+1)/16)
1092 : /* Map errors 1:1 up to +- MAXJSAMPLE/16 */
1093 0 : out = 0;
1094 0 : for (in = 0; in < STEPSIZE; in++, out++) {
1095 0 : table[in] = out; table[-in] = -out;
1096 : }
1097 : /* Map errors 1:2 up to +- 3*MAXJSAMPLE/16 */
1098 0 : for (; in < STEPSIZE*3; in++, out += (in&1) ? 0 : 1) {
1099 0 : table[in] = out; table[-in] = -out;
1100 : }
1101 : /* Clamp the rest to final out value (which is (MAXJSAMPLE+1)/8) */
1102 0 : for (; in <= MAXJSAMPLE; in++) {
1103 0 : table[in] = out; table[-in] = -out;
1104 : }
1105 : #undef STEPSIZE
1106 0 : }
1107 :
1108 :
1109 : /*
1110 : * Finish up at the end of each pass.
1111 : */
1112 :
1113 : METHODDEF(void)
1114 0 : finish_pass1 (j_decompress_ptr cinfo)
1115 : {
1116 0 : my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
1117 :
1118 : /* Select the representative colors and fill in cinfo->colormap */
1119 0 : cinfo->colormap = cquantize->sv_colormap;
1120 0 : select_colors(cinfo, cquantize->desired);
1121 : /* Force next pass to zero the color index table */
1122 0 : cquantize->needs_zeroed = TRUE;
1123 0 : }
1124 :
1125 :
1126 : METHODDEF(void)
1127 0 : finish_pass2 (j_decompress_ptr cinfo)
1128 : {
1129 : /* no work */
1130 0 : }
1131 :
1132 :
1133 : /*
1134 : * Initialize for each processing pass.
1135 : */
1136 :
1137 : METHODDEF(void)
1138 0 : start_pass_2_quant (j_decompress_ptr cinfo, boolean is_pre_scan)
1139 : {
1140 0 : my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
1141 0 : hist3d histogram = cquantize->histogram;
1142 : int i;
1143 :
1144 : /* Only F-S dithering or no dithering is supported. */
1145 : /* If user asks for ordered dither, give him F-S. */
1146 0 : if (cinfo->dither_mode != JDITHER_NONE)
1147 0 : cinfo->dither_mode = JDITHER_FS;
1148 :
1149 0 : if (is_pre_scan) {
1150 : /* Set up method pointers */
1151 0 : cquantize->pub.color_quantize = prescan_quantize;
1152 0 : cquantize->pub.finish_pass = finish_pass1;
1153 0 : cquantize->needs_zeroed = TRUE; /* Always zero histogram */
1154 : } else {
1155 : /* Set up method pointers */
1156 0 : if (cinfo->dither_mode == JDITHER_FS)
1157 0 : cquantize->pub.color_quantize = pass2_fs_dither;
1158 : else
1159 0 : cquantize->pub.color_quantize = pass2_no_dither;
1160 0 : cquantize->pub.finish_pass = finish_pass2;
1161 :
1162 : /* Make sure color count is acceptable */
1163 0 : i = cinfo->actual_number_of_colors;
1164 0 : if (i < 1)
1165 0 : ERREXIT1(cinfo, JERR_QUANT_FEW_COLORS, 1);
1166 0 : if (i > MAXNUMCOLORS)
1167 0 : ERREXIT1(cinfo, JERR_QUANT_MANY_COLORS, MAXNUMCOLORS);
1168 :
1169 0 : if (cinfo->dither_mode == JDITHER_FS) {
1170 0 : size_t arraysize = (size_t) ((cinfo->output_width + 2) *
1171 : (3 * sizeof(FSERROR)));
1172 : /* Allocate Floyd-Steinberg workspace if we didn't already. */
1173 0 : if (cquantize->fserrors == NULL)
1174 0 : cquantize->fserrors = (FSERRPTR) (*cinfo->mem->alloc_large)
1175 : ((j_common_ptr) cinfo, JPOOL_IMAGE, arraysize);
1176 : /* Initialize the propagated errors to zero. */
1177 0 : jzero_far((void *) cquantize->fserrors, arraysize);
1178 : /* Make the error-limit table if we didn't already. */
1179 0 : if (cquantize->error_limiter == NULL)
1180 0 : init_error_limit(cinfo);
1181 0 : cquantize->on_odd_row = FALSE;
1182 : }
1183 :
1184 : }
1185 : /* Zero the histogram or inverse color map, if necessary */
1186 0 : if (cquantize->needs_zeroed) {
1187 0 : for (i = 0; i < HIST_C0_ELEMS; i++) {
1188 0 : jzero_far((void *) histogram[i],
1189 : HIST_C1_ELEMS*HIST_C2_ELEMS * sizeof(histcell));
1190 : }
1191 0 : cquantize->needs_zeroed = FALSE;
1192 : }
1193 0 : }
1194 :
1195 :
1196 : /*
1197 : * Switch to a new external colormap between output passes.
1198 : */
1199 :
1200 : METHODDEF(void)
1201 0 : new_color_map_2_quant (j_decompress_ptr cinfo)
1202 : {
1203 0 : my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
1204 :
1205 : /* Reset the inverse color map */
1206 0 : cquantize->needs_zeroed = TRUE;
1207 0 : }
1208 :
1209 :
1210 : /*
1211 : * Module initialization routine for 2-pass color quantization.
1212 : */
1213 :
1214 : GLOBAL(void)
1215 0 : jinit_2pass_quantizer (j_decompress_ptr cinfo)
1216 : {
1217 : my_cquantize_ptr cquantize;
1218 : int i;
1219 :
1220 0 : cquantize = (my_cquantize_ptr)
1221 0 : (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
1222 : sizeof(my_cquantizer));
1223 0 : cinfo->cquantize = (struct jpeg_color_quantizer *) cquantize;
1224 0 : cquantize->pub.start_pass = start_pass_2_quant;
1225 0 : cquantize->pub.new_color_map = new_color_map_2_quant;
1226 0 : cquantize->fserrors = NULL; /* flag optional arrays not allocated */
1227 0 : cquantize->error_limiter = NULL;
1228 :
1229 : /* Make sure jdmaster didn't give me a case I can't handle */
1230 0 : if (cinfo->out_color_components != 3)
1231 0 : ERREXIT(cinfo, JERR_NOTIMPL);
1232 :
1233 : /* Allocate the histogram/inverse colormap storage */
1234 0 : cquantize->histogram = (hist3d) (*cinfo->mem->alloc_small)
1235 : ((j_common_ptr) cinfo, JPOOL_IMAGE, HIST_C0_ELEMS * sizeof(hist2d));
1236 0 : for (i = 0; i < HIST_C0_ELEMS; i++) {
1237 0 : cquantize->histogram[i] = (hist2d) (*cinfo->mem->alloc_large)
1238 : ((j_common_ptr) cinfo, JPOOL_IMAGE,
1239 : HIST_C1_ELEMS*HIST_C2_ELEMS * sizeof(histcell));
1240 : }
1241 0 : cquantize->needs_zeroed = TRUE; /* histogram is garbage now */
1242 :
1243 : /* Allocate storage for the completed colormap, if required.
1244 : * We do this now since it may affect the memory manager's space
1245 : * calculations.
1246 : */
1247 0 : if (cinfo->enable_2pass_quant) {
1248 : /* Make sure color count is acceptable */
1249 0 : int desired = cinfo->desired_number_of_colors;
1250 : /* Lower bound on # of colors ... somewhat arbitrary as long as > 0 */
1251 0 : if (desired < 8)
1252 0 : ERREXIT1(cinfo, JERR_QUANT_FEW_COLORS, 8);
1253 : /* Make sure colormap indexes can be represented by JSAMPLEs */
1254 0 : if (desired > MAXNUMCOLORS)
1255 0 : ERREXIT1(cinfo, JERR_QUANT_MANY_COLORS, MAXNUMCOLORS);
1256 0 : cquantize->sv_colormap = (*cinfo->mem->alloc_sarray)
1257 : ((j_common_ptr) cinfo,JPOOL_IMAGE, (JDIMENSION) desired, (JDIMENSION) 3);
1258 0 : cquantize->desired = desired;
1259 : } else
1260 0 : cquantize->sv_colormap = NULL;
1261 :
1262 : /* Only F-S dithering or no dithering is supported. */
1263 : /* If user asks for ordered dither, give him F-S. */
1264 0 : if (cinfo->dither_mode != JDITHER_NONE)
1265 0 : cinfo->dither_mode = JDITHER_FS;
1266 :
1267 : /* Allocate Floyd-Steinberg workspace if necessary.
1268 : * This isn't really needed until pass 2, but again it may affect the memory
1269 : * manager's space calculations. Although we will cope with a later change
1270 : * in dither_mode, we do not promise to honor max_memory_to_use if
1271 : * dither_mode changes.
1272 : */
1273 0 : if (cinfo->dither_mode == JDITHER_FS) {
1274 0 : cquantize->fserrors = (FSERRPTR) (*cinfo->mem->alloc_large)
1275 : ((j_common_ptr) cinfo, JPOOL_IMAGE,
1276 0 : (size_t) ((cinfo->output_width + 2) * (3 * sizeof(FSERROR))));
1277 : /* Might as well create the error-limiting table too. */
1278 0 : init_error_limit(cinfo);
1279 : }
1280 0 : }
1281 :
1282 : #endif /* QUANT_2PASS_SUPPORTED */
|
