]>
Commit | Line | Data |
---|---|---|
1 | ///////////////////////////////////////////////////////////////////////////// | |
2 | // Name: quantize.cpp | |
3 | // Purpose: wxQuantize implementation | |
4 | // Author: Julian Smart | |
5 | // Modified by: | |
6 | // Created: 22/6/2000 | |
7 | // RCS-ID: $Id$ | |
8 | // Copyright: (c) Thomas G. Lane, Vaclav Slavik, Julian Smart | |
9 | // Licence: wxWindows licence + JPEG library licence | |
10 | ///////////////////////////////////////////////////////////////////////////// | |
11 | ||
12 | /* | |
13 | * jquant2.c | |
14 | * | |
15 | * Copyright (C) 1991-1996, Thomas G. Lane. | |
16 | * This file is part of the Independent JPEG Group's software. | |
17 | * For conditions of distribution and use, see the accompanying README file. | |
18 | * | |
19 | * This file contains 2-pass color quantization (color mapping) routines. | |
20 | * These routines provide selection of a custom color map for an image, | |
21 | * followed by mapping of the image to that color map, with optional | |
22 | * Floyd-Steinberg dithering. | |
23 | * It is also possible to use just the second pass to map to an arbitrary | |
24 | * externally-given color map. | |
25 | * | |
26 | * Note: ordered dithering is not supported, since there isn't any fast | |
27 | * way to compute intercolor distances; it's unclear that ordered dither's | |
28 | * fundamental assumptions even hold with an irregularly spaced color map. | |
29 | */ | |
30 | ||
31 | /* modified by Vaclav Slavik for use as jpeglib-independent module */ | |
32 | ||
33 | #ifdef __GNUG__ | |
34 | #pragma implementation "quantize.h" | |
35 | #endif | |
36 | ||
37 | // For compilers that support precompilation, includes "wx/wx.h". | |
38 | #include "wx/wxprec.h" | |
39 | ||
40 | #ifdef __BORLANDC__ | |
41 | #pragma hdrstop | |
42 | #endif | |
43 | ||
44 | #ifndef WX_PRECOMP | |
45 | #include "wx/wx.h" | |
46 | #endif | |
47 | ||
48 | #include "wx/image.h" | |
49 | #include "wx/quantize.h" | |
50 | ||
51 | #ifdef __WXMSW__ | |
52 | #include <windows.h> | |
53 | #endif | |
54 | ||
55 | #include <stdlib.h> | |
56 | #include <string.h> | |
57 | ||
58 | #define RGB_RED 0 | |
59 | #define RGB_GREEN 1 | |
60 | #define RGB_BLUE 2 | |
61 | #define RGB_PIXELSIZE 3 | |
62 | ||
63 | #define MAXJSAMPLE 255 | |
64 | #define CENTERJSAMPLE 128 | |
65 | #define BITS_IN_JSAMPLE 8 | |
66 | #define GETJSAMPLE(value) ((int) (value)) | |
67 | ||
68 | #define RIGHT_SHIFT(x,shft) ((x) >> (shft)) | |
69 | ||
70 | typedef unsigned short UINT16; | |
71 | typedef signed short INT16; | |
72 | typedef signed int INT32; | |
73 | ||
74 | typedef unsigned char JSAMPLE; | |
75 | typedef JSAMPLE *JSAMPROW; | |
76 | typedef JSAMPROW *JSAMPARRAY; | |
77 | typedef unsigned int JDIMENSION; | |
78 | ||
79 | typedef struct { | |
80 | void *cquantize; | |
81 | JDIMENSION output_width; | |
82 | JSAMPARRAY colormap; | |
83 | int actual_number_of_colors; | |
84 | int desired_number_of_colors; | |
85 | JSAMPLE *sample_range_limit, *srl_orig; | |
86 | } j_decompress; | |
87 | ||
88 | typedef j_decompress *j_decompress_ptr; | |
89 | ||
90 | ||
91 | /* | |
92 | * This module implements the well-known Heckbert paradigm for color | |
93 | * quantization. Most of the ideas used here can be traced back to | |
94 | * Heckbert's seminal paper | |
95 | * Heckbert, Paul. "Color Image Quantization for Frame Buffer Display", | |
96 | * Proc. SIGGRAPH '82, Computer Graphics v.16 #3 (July 1982), pp 297-304. | |
97 | * | |
98 | * In the first pass over the image, we accumulate a histogram showing the | |
99 | * usage count of each possible color. To keep the histogram to a reasonable | |
100 | * size, we reduce the precision of the input; typical practice is to retain | |
101 | * 5 or 6 bits per color, so that 8 or 4 different input values are counted | |
102 | * in the same histogram cell. | |
103 | * | |
104 | * Next, the color-selection step begins with a box representing the whole | |
105 | * color space, and repeatedly splits the "largest" remaining box until we | |
106 | * have as many boxes as desired colors. Then the mean color in each | |
107 | * remaining box becomes one of the possible output colors. | |
108 | * | |
109 | * The second pass over the image maps each input pixel to the closest output | |
110 | * color (optionally after applying a Floyd-Steinberg dithering correction). | |
111 | * This mapping is logically trivial, but making it go fast enough requires | |
112 | * considerable care. | |
113 | * | |
114 | * Heckbert-style quantizers vary a good deal in their policies for choosing | |
115 | * the "largest" box and deciding where to cut it. The particular policies | |
116 | * used here have proved out well in experimental comparisons, but better ones | |
117 | * may yet be found. | |
118 | * | |
119 | * In earlier versions of the IJG code, this module quantized in YCbCr color | |
120 | * space, processing the raw upsampled data without a color conversion step. | |
121 | * This allowed the color conversion math to be done only once per colormap | |
122 | * entry, not once per pixel. However, that optimization precluded other | |
123 | * useful optimizations (such as merging color conversion with upsampling) | |
124 | * and it also interfered with desired capabilities such as quantizing to an | |
125 | * externally-supplied colormap. We have therefore abandoned that approach. | |
126 | * The present code works in the post-conversion color space, typically RGB. | |
127 | * | |
128 | * To improve the visual quality of the results, we actually work in scaled | |
129 | * RGB space, giving G distances more weight than R, and R in turn more than | |
130 | * B. To do everything in integer math, we must use integer scale factors. | |
131 | * The 2/3/1 scale factors used here correspond loosely to the relative | |
132 | * weights of the colors in the NTSC grayscale equation. | |
133 | * If you want to use this code to quantize a non-RGB color space, you'll | |
134 | * probably need to change these scale factors. | |
135 | */ | |
136 | ||
137 | #define R_SCALE 2 /* scale R distances by this much */ | |
138 | #define G_SCALE 3 /* scale G distances by this much */ | |
139 | #define B_SCALE 1 /* and B by this much */ | |
140 | ||
141 | /* Relabel R/G/B as components 0/1/2, respecting the RGB ordering defined | |
142 | * in jmorecfg.h. As the code stands, it will do the right thing for R,G,B | |
143 | * and B,G,R orders. If you define some other weird order in jmorecfg.h, | |
144 | * you'll get compile errors until you extend this logic. In that case | |
145 | * you'll probably want to tweak the histogram sizes too. | |
146 | */ | |
147 | ||
148 | #if RGB_RED == 0 | |
149 | #define C0_SCALE R_SCALE | |
150 | #endif | |
151 | #if RGB_BLUE == 0 | |
152 | #define C0_SCALE B_SCALE | |
153 | #endif | |
154 | #if RGB_GREEN == 1 | |
155 | #define C1_SCALE G_SCALE | |
156 | #endif | |
157 | #if RGB_RED == 2 | |
158 | #define C2_SCALE R_SCALE | |
159 | #endif | |
160 | #if RGB_BLUE == 2 | |
161 | #define C2_SCALE B_SCALE | |
162 | #endif | |
163 | ||
164 | ||
165 | /* | |
166 | * First we have the histogram data structure and routines for creating it. | |
167 | * | |
168 | * The number of bits of precision can be adjusted by changing these symbols. | |
169 | * We recommend keeping 6 bits for G and 5 each for R and B. | |
170 | * If you have plenty of memory and cycles, 6 bits all around gives marginally | |
171 | * better results; if you are short of memory, 5 bits all around will save | |
172 | * some space but degrade the results. | |
173 | * To maintain a fully accurate histogram, we'd need to allocate a "long" | |
174 | * (preferably unsigned long) for each cell. In practice this is overkill; | |
175 | * we can get by with 16 bits per cell. Few of the cell counts will overflow, | |
176 | * and clamping those that do overflow to the maximum value will give close- | |
177 | * enough results. This reduces the recommended histogram size from 256Kb | |
178 | * to 128Kb, which is a useful savings on PC-class machines. | |
179 | * (In the second pass the histogram space is re-used for pixel mapping data; | |
180 | * in that capacity, each cell must be able to store zero to the number of | |
181 | * desired colors. 16 bits/cell is plenty for that too.) | |
182 | * Since the JPEG code is intended to run in small memory model on 80x86 | |
183 | * machines, we can't just allocate the histogram in one chunk. Instead | |
184 | * of a true 3-D array, we use a row of pointers to 2-D arrays. Each | |
185 | * pointer corresponds to a C0 value (typically 2^5 = 32 pointers) and | |
186 | * each 2-D array has 2^6*2^5 = 2048 or 2^6*2^6 = 4096 entries. Note that | |
187 | * on 80x86 machines, the pointer row is in near memory but the actual | |
188 | * arrays are in far memory (same arrangement as we use for image arrays). | |
189 | */ | |
190 | ||
191 | #define MAXNUMCOLORS (MAXJSAMPLE+1) /* maximum size of colormap */ | |
192 | ||
193 | /* These will do the right thing for either R,G,B or B,G,R color order, | |
194 | * but you may not like the results for other color orders. | |
195 | */ | |
196 | #define HIST_C0_BITS 5 /* bits of precision in R/B histogram */ | |
197 | #define HIST_C1_BITS 6 /* bits of precision in G histogram */ | |
198 | #define HIST_C2_BITS 5 /* bits of precision in B/R histogram */ | |
199 | ||
200 | /* Number of elements along histogram axes. */ | |
201 | #define HIST_C0_ELEMS (1<<HIST_C0_BITS) | |
202 | #define HIST_C1_ELEMS (1<<HIST_C1_BITS) | |
203 | #define HIST_C2_ELEMS (1<<HIST_C2_BITS) | |
204 | ||
205 | /* These are the amounts to shift an input value to get a histogram index. */ | |
206 | #define C0_SHIFT (BITS_IN_JSAMPLE-HIST_C0_BITS) | |
207 | #define C1_SHIFT (BITS_IN_JSAMPLE-HIST_C1_BITS) | |
208 | #define C2_SHIFT (BITS_IN_JSAMPLE-HIST_C2_BITS) | |
209 | ||
210 | ||
211 | typedef UINT16 histcell; /* histogram cell; prefer an unsigned type */ | |
212 | ||
213 | typedef histcell * histptr; /* for pointers to histogram cells */ | |
214 | ||
215 | typedef histcell hist1d[HIST_C2_ELEMS]; /* typedefs for the array */ | |
216 | typedef hist1d * hist2d; /* type for the 2nd-level pointers */ | |
217 | typedef hist2d * hist3d; /* type for top-level pointer */ | |
218 | ||
219 | ||
220 | /* Declarations for Floyd-Steinberg dithering. | |
221 | * | |
222 | * Errors are accumulated into the array fserrors[], at a resolution of | |
223 | * 1/16th of a pixel count. The error at a given pixel is propagated | |
224 | * to its not-yet-processed neighbors using the standard F-S fractions, | |
225 | * ... (here) 7/16 | |
226 | * 3/16 5/16 1/16 | |
227 | * We work left-to-right on even rows, right-to-left on odd rows. | |
228 | * | |
229 | * We can get away with a single array (holding one row's worth of errors) | |
230 | * by using it to store the current row's errors at pixel columns not yet | |
231 | * processed, but the next row's errors at columns already processed. We | |
232 | * need only a few extra variables to hold the errors immediately around the | |
233 | * current column. (If we are lucky, those variables are in registers, but | |
234 | * even if not, they're probably cheaper to access than array elements are.) | |
235 | * | |
236 | * The fserrors[] array has (#columns + 2) entries; the extra entry at | |
237 | * each end saves us from special-casing the first and last pixels. | |
238 | * Each entry is three values long, one value for each color component. | |
239 | * | |
240 | * Note: on a wide image, we might not have enough room in a PC's near data | |
241 | * segment to hold the error array; so it is allocated with alloc_large. | |
242 | */ | |
243 | ||
244 | #if BITS_IN_JSAMPLE == 8 | |
245 | typedef INT16 FSERROR; /* 16 bits should be enough */ | |
246 | typedef int LOCFSERROR; /* use 'int' for calculation temps */ | |
247 | #else | |
248 | typedef INT32 FSERROR; /* may need more than 16 bits */ | |
249 | typedef INT32 LOCFSERROR; /* be sure calculation temps are big enough */ | |
250 | #endif | |
251 | ||
252 | typedef FSERROR *FSERRPTR; /* pointer to error array (in storage!) */ | |
253 | ||
254 | ||
255 | /* Private subobject */ | |
256 | ||
257 | typedef struct { | |
258 | ||
259 | struct { | |
260 | void (*finish_pass)(j_decompress_ptr); | |
261 | void (*color_quantize)(j_decompress_ptr, JSAMPARRAY, JSAMPARRAY, int); | |
262 | void (*start_pass)(j_decompress_ptr, bool); | |
263 | void (*new_color_map)(j_decompress_ptr); | |
264 | } pub; | |
265 | ||
266 | /* Space for the eventually created colormap is stashed here */ | |
267 | JSAMPARRAY sv_colormap; /* colormap allocated at init time */ | |
268 | int desired; /* desired # of colors = size of colormap */ | |
269 | ||
270 | /* Variables for accumulating image statistics */ | |
271 | hist3d histogram; /* pointer to the histogram */ | |
272 | ||
273 | bool needs_zeroed; /* true if next pass must zero histogram */ | |
274 | ||
275 | /* Variables for Floyd-Steinberg dithering */ | |
276 | FSERRPTR fserrors; /* accumulated errors */ | |
277 | bool on_odd_row; /* flag to remember which row we are on */ | |
278 | int * error_limiter; /* table for clamping the applied error */ | |
279 | } my_cquantizer; | |
280 | ||
281 | typedef my_cquantizer * my_cquantize_ptr; | |
282 | ||
283 | ||
284 | /* | |
285 | * Prescan some rows of pixels. | |
286 | * In this module the prescan simply updates the histogram, which has been | |
287 | * initialized to zeroes by start_pass. | |
288 | * An output_buf parameter is required by the method signature, but no data | |
289 | * is actually output (in fact the buffer controller is probably passing a | |
290 | * NULL pointer). | |
291 | */ | |
292 | ||
293 | void | |
294 | prescan_quantize (j_decompress_ptr cinfo, JSAMPARRAY input_buf, | |
295 | JSAMPARRAY output_buf, int num_rows) | |
296 | { | |
297 | my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize; | |
298 | register JSAMPROW ptr; | |
299 | register histptr histp; | |
300 | register hist3d histogram = cquantize->histogram; | |
301 | int row; | |
302 | JDIMENSION col; | |
303 | JDIMENSION width = cinfo->output_width; | |
304 | ||
305 | for (row = 0; row < num_rows; row++) { | |
306 | ptr = input_buf[row]; | |
307 | for (col = width; col > 0; col--) { | |
308 | ||
309 | { | |
310 | ||
311 | /* get pixel value and index into the histogram */ | |
312 | histp = & histogram[GETJSAMPLE(ptr[0]) >> C0_SHIFT] | |
313 | [GETJSAMPLE(ptr[1]) >> C1_SHIFT] | |
314 | [GETJSAMPLE(ptr[2]) >> C2_SHIFT]; | |
315 | /* increment, check for overflow and undo increment if so. */ | |
316 | if (++(*histp) <= 0) | |
317 | (*histp)--; | |
318 | } | |
319 | ptr += 3; | |
320 | } | |
321 | } | |
322 | } | |
323 | ||
324 | ||
325 | /* | |
326 | * Next we have the really interesting routines: selection of a colormap | |
327 | * given the completed histogram. | |
328 | * These routines work with a list of "boxes", each representing a rectangular | |
329 | * subset of the input color space (to histogram precision). | |
330 | */ | |
331 | ||
332 | typedef struct { | |
333 | /* The bounds of the box (inclusive); expressed as histogram indexes */ | |
334 | int c0min, c0max; | |
335 | int c1min, c1max; | |
336 | int c2min, c2max; | |
337 | /* The volume (actually 2-norm) of the box */ | |
338 | INT32 volume; | |
339 | /* The number of nonzero histogram cells within this box */ | |
340 | long colorcount; | |
341 | } box; | |
342 | ||
343 | typedef box * boxptr; | |
344 | ||
345 | ||
346 | boxptr | |
347 | find_biggest_color_pop (boxptr boxlist, int numboxes) | |
348 | /* Find the splittable box with the largest color population */ | |
349 | /* Returns NULL if no splittable boxes remain */ | |
350 | { | |
351 | register boxptr boxp; | |
352 | register int i; | |
353 | register long maxc = 0; | |
354 | boxptr which = NULL; | |
355 | ||
356 | for (i = 0, boxp = boxlist; i < numboxes; i++, boxp++) { | |
357 | if (boxp->colorcount > maxc && boxp->volume > 0) { | |
358 | which = boxp; | |
359 | maxc = boxp->colorcount; | |
360 | } | |
361 | } | |
362 | return which; | |
363 | } | |
364 | ||
365 | ||
366 | boxptr | |
367 | find_biggest_volume (boxptr boxlist, int numboxes) | |
368 | /* Find the splittable box with the largest (scaled) volume */ | |
369 | /* Returns NULL if no splittable boxes remain */ | |
370 | { | |
371 | register boxptr boxp; | |
372 | register int i; | |
373 | register INT32 maxv = 0; | |
374 | boxptr which = NULL; | |
375 | ||
376 | for (i = 0, boxp = boxlist; i < numboxes; i++, boxp++) { | |
377 | if (boxp->volume > maxv) { | |
378 | which = boxp; | |
379 | maxv = boxp->volume; | |
380 | } | |
381 | } | |
382 | return which; | |
383 | } | |
384 | ||
385 | ||
386 | void | |
387 | update_box (j_decompress_ptr cinfo, boxptr boxp) | |
388 | /* Shrink the min/max bounds of a box to enclose only nonzero elements, */ | |
389 | /* and recompute its volume and population */ | |
390 | { | |
391 | my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize; | |
392 | hist3d histogram = cquantize->histogram; | |
393 | histptr histp; | |
394 | int c0,c1,c2; | |
395 | int c0min,c0max,c1min,c1max,c2min,c2max; | |
396 | INT32 dist0,dist1,dist2; | |
397 | long ccount; | |
398 | ||
399 | c0min = boxp->c0min; c0max = boxp->c0max; | |
400 | c1min = boxp->c1min; c1max = boxp->c1max; | |
401 | c2min = boxp->c2min; c2max = boxp->c2max; | |
402 | ||
403 | if (c0max > c0min) | |
404 | for (c0 = c0min; c0 <= c0max; c0++) | |
405 | for (c1 = c1min; c1 <= c1max; c1++) { | |
406 | histp = & histogram[c0][c1][c2min]; | |
407 | for (c2 = c2min; c2 <= c2max; c2++) | |
408 | if (*histp++ != 0) { | |
409 | boxp->c0min = c0min = c0; | |
410 | goto have_c0min; | |
411 | } | |
412 | } | |
413 | have_c0min: | |
414 | if (c0max > c0min) | |
415 | for (c0 = c0max; c0 >= c0min; c0--) | |
416 | for (c1 = c1min; c1 <= c1max; c1++) { | |
417 | histp = & histogram[c0][c1][c2min]; | |
418 | for (c2 = c2min; c2 <= c2max; c2++) | |
419 | if (*histp++ != 0) { | |
420 | boxp->c0max = c0max = c0; | |
421 | goto have_c0max; | |
422 | } | |
423 | } | |
424 | have_c0max: | |
425 | if (c1max > c1min) | |
426 | for (c1 = c1min; c1 <= c1max; c1++) | |
427 | for (c0 = c0min; c0 <= c0max; c0++) { | |
428 | histp = & histogram[c0][c1][c2min]; | |
429 | for (c2 = c2min; c2 <= c2max; c2++) | |
430 | if (*histp++ != 0) { | |
431 | boxp->c1min = c1min = c1; | |
432 | goto have_c1min; | |
433 | } | |
434 | } | |
435 | have_c1min: | |
436 | if (c1max > c1min) | |
437 | for (c1 = c1max; c1 >= c1min; c1--) | |
438 | for (c0 = c0min; c0 <= c0max; c0++) { | |
439 | histp = & histogram[c0][c1][c2min]; | |
440 | for (c2 = c2min; c2 <= c2max; c2++) | |
441 | if (*histp++ != 0) { | |
442 | boxp->c1max = c1max = c1; | |
443 | goto have_c1max; | |
444 | } | |
445 | } | |
446 | have_c1max: | |
447 | if (c2max > c2min) | |
448 | for (c2 = c2min; c2 <= c2max; c2++) | |
449 | for (c0 = c0min; c0 <= c0max; c0++) { | |
450 | histp = & histogram[c0][c1min][c2]; | |
451 | for (c1 = c1min; c1 <= c1max; c1++, histp += HIST_C2_ELEMS) | |
452 | if (*histp != 0) { | |
453 | boxp->c2min = c2min = c2; | |
454 | goto have_c2min; | |
455 | } | |
456 | } | |
457 | have_c2min: | |
458 | if (c2max > c2min) | |
459 | for (c2 = c2max; c2 >= c2min; c2--) | |
460 | for (c0 = c0min; c0 <= c0max; c0++) { | |
461 | histp = & histogram[c0][c1min][c2]; | |
462 | for (c1 = c1min; c1 <= c1max; c1++, histp += HIST_C2_ELEMS) | |
463 | if (*histp != 0) { | |
464 | boxp->c2max = c2max = c2; | |
465 | goto have_c2max; | |
466 | } | |
467 | } | |
468 | have_c2max: | |
469 | ||
470 | /* Update box volume. | |
471 | * We use 2-norm rather than real volume here; this biases the method | |
472 | * against making long narrow boxes, and it has the side benefit that | |
473 | * a box is splittable iff norm > 0. | |
474 | * Since the differences are expressed in histogram-cell units, | |
475 | * we have to shift back to JSAMPLE units to get consistent distances; | |
476 | * after which, we scale according to the selected distance scale factors. | |
477 | */ | |
478 | dist0 = ((c0max - c0min) << C0_SHIFT) * C0_SCALE; | |
479 | dist1 = ((c1max - c1min) << C1_SHIFT) * C1_SCALE; | |
480 | dist2 = ((c2max - c2min) << C2_SHIFT) * C2_SCALE; | |
481 | boxp->volume = dist0*dist0 + dist1*dist1 + dist2*dist2; | |
482 | ||
483 | /* Now scan remaining volume of box and compute population */ | |
484 | ccount = 0; | |
485 | for (c0 = c0min; c0 <= c0max; c0++) | |
486 | for (c1 = c1min; c1 <= c1max; c1++) { | |
487 | histp = & histogram[c0][c1][c2min]; | |
488 | for (c2 = c2min; c2 <= c2max; c2++, histp++) | |
489 | if (*histp != 0) { | |
490 | ccount++; | |
491 | } | |
492 | } | |
493 | boxp->colorcount = ccount; | |
494 | } | |
495 | ||
496 | ||
497 | int | |
498 | median_cut (j_decompress_ptr cinfo, boxptr boxlist, int numboxes, | |
499 | int desired_colors) | |
500 | /* Repeatedly select and split the largest box until we have enough boxes */ | |
501 | { | |
502 | int n,lb; | |
503 | int c0,c1,c2,cmax; | |
504 | register boxptr b1,b2; | |
505 | ||
506 | while (numboxes < desired_colors) { | |
507 | /* Select box to split. | |
508 | * Current algorithm: by population for first half, then by volume. | |
509 | */ | |
510 | if (numboxes*2 <= desired_colors) { | |
511 | b1 = find_biggest_color_pop(boxlist, numboxes); | |
512 | } else { | |
513 | b1 = find_biggest_volume(boxlist, numboxes); | |
514 | } | |
515 | if (b1 == NULL) /* no splittable boxes left! */ | |
516 | break; | |
517 | b2 = &boxlist[numboxes]; /* where new box will go */ | |
518 | /* Copy the color bounds to the new box. */ | |
519 | b2->c0max = b1->c0max; b2->c1max = b1->c1max; b2->c2max = b1->c2max; | |
520 | b2->c0min = b1->c0min; b2->c1min = b1->c1min; b2->c2min = b1->c2min; | |
521 | /* Choose which axis to split the box on. | |
522 | * Current algorithm: longest scaled axis. | |
523 | * See notes in update_box about scaling distances. | |
524 | */ | |
525 | c0 = ((b1->c0max - b1->c0min) << C0_SHIFT) * C0_SCALE; | |
526 | c1 = ((b1->c1max - b1->c1min) << C1_SHIFT) * C1_SCALE; | |
527 | c2 = ((b1->c2max - b1->c2min) << C2_SHIFT) * C2_SCALE; | |
528 | /* We want to break any ties in favor of green, then red, blue last. | |
529 | * This code does the right thing for R,G,B or B,G,R color orders only. | |
530 | */ | |
531 | #if RGB_RED == 0 | |
532 | cmax = c1; n = 1; | |
533 | if (c0 > cmax) { cmax = c0; n = 0; } | |
534 | if (c2 > cmax) { n = 2; } | |
535 | #else | |
536 | cmax = c1; n = 1; | |
537 | if (c2 > cmax) { cmax = c2; n = 2; } | |
538 | if (c0 > cmax) { n = 0; } | |
539 | #endif | |
540 | /* Choose split point along selected axis, and update box bounds. | |
541 | * Current algorithm: split at halfway point. | |
542 | * (Since the box has been shrunk to minimum volume, | |
543 | * any split will produce two nonempty subboxes.) | |
544 | * Note that lb value is max for lower box, so must be < old max. | |
545 | */ | |
546 | switch (n) { | |
547 | case 0: | |
548 | lb = (b1->c0max + b1->c0min) / 2; | |
549 | b1->c0max = lb; | |
550 | b2->c0min = lb+1; | |
551 | break; | |
552 | case 1: | |
553 | lb = (b1->c1max + b1->c1min) / 2; | |
554 | b1->c1max = lb; | |
555 | b2->c1min = lb+1; | |
556 | break; | |
557 | case 2: | |
558 | lb = (b1->c2max + b1->c2min) / 2; | |
559 | b1->c2max = lb; | |
560 | b2->c2min = lb+1; | |
561 | break; | |
562 | } | |
563 | /* Update stats for boxes */ | |
564 | update_box(cinfo, b1); | |
565 | update_box(cinfo, b2); | |
566 | numboxes++; | |
567 | } | |
568 | return numboxes; | |
569 | } | |
570 | ||
571 | ||
572 | void | |
573 | compute_color (j_decompress_ptr cinfo, boxptr boxp, int icolor) | |
574 | /* Compute representative color for a box, put it in colormap[icolor] */ | |
575 | { | |
576 | /* Current algorithm: mean weighted by pixels (not colors) */ | |
577 | /* Note it is important to get the rounding correct! */ | |
578 | my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize; | |
579 | hist3d histogram = cquantize->histogram; | |
580 | histptr histp; | |
581 | int c0,c1,c2; | |
582 | int c0min,c0max,c1min,c1max,c2min,c2max; | |
583 | long count; | |
584 | long total = 0; | |
585 | long c0total = 0; | |
586 | long c1total = 0; | |
587 | long c2total = 0; | |
588 | ||
589 | c0min = boxp->c0min; c0max = boxp->c0max; | |
590 | c1min = boxp->c1min; c1max = boxp->c1max; | |
591 | c2min = boxp->c2min; c2max = boxp->c2max; | |
592 | ||
593 | for (c0 = c0min; c0 <= c0max; c0++) | |
594 | for (c1 = c1min; c1 <= c1max; c1++) { | |
595 | histp = & histogram[c0][c1][c2min]; | |
596 | for (c2 = c2min; c2 <= c2max; c2++) { | |
597 | if ((count = *histp++) != 0) { | |
598 | total += count; | |
599 | c0total += ((c0 << C0_SHIFT) + ((1<<C0_SHIFT)>>1)) * count; | |
600 | c1total += ((c1 << C1_SHIFT) + ((1<<C1_SHIFT)>>1)) * count; | |
601 | c2total += ((c2 << C2_SHIFT) + ((1<<C2_SHIFT)>>1)) * count; | |
602 | } | |
603 | } | |
604 | } | |
605 | ||
606 | cinfo->colormap[0][icolor] = (JSAMPLE) ((c0total + (total>>1)) / total); | |
607 | cinfo->colormap[1][icolor] = (JSAMPLE) ((c1total + (total>>1)) / total); | |
608 | cinfo->colormap[2][icolor] = (JSAMPLE) ((c2total + (total>>1)) / total); | |
609 | } | |
610 | ||
611 | ||
612 | static void | |
613 | select_colors (j_decompress_ptr cinfo, int desired_colors) | |
614 | /* Master routine for color selection */ | |
615 | { | |
616 | boxptr boxlist; | |
617 | int numboxes; | |
618 | int i; | |
619 | ||
620 | /* Allocate workspace for box list */ | |
621 | boxlist = (boxptr) malloc(desired_colors * sizeof(box)); | |
622 | /* Initialize one box containing whole space */ | |
623 | numboxes = 1; | |
624 | boxlist[0].c0min = 0; | |
625 | boxlist[0].c0max = MAXJSAMPLE >> C0_SHIFT; | |
626 | boxlist[0].c1min = 0; | |
627 | boxlist[0].c1max = MAXJSAMPLE >> C1_SHIFT; | |
628 | boxlist[0].c2min = 0; | |
629 | boxlist[0].c2max = MAXJSAMPLE >> C2_SHIFT; | |
630 | /* Shrink it to actually-used volume and set its statistics */ | |
631 | update_box(cinfo, & boxlist[0]); | |
632 | /* Perform median-cut to produce final box list */ | |
633 | numboxes = median_cut(cinfo, boxlist, numboxes, desired_colors); | |
634 | /* Compute the representative color for each box, fill colormap */ | |
635 | for (i = 0; i < numboxes; i++) | |
636 | compute_color(cinfo, & boxlist[i], i); | |
637 | cinfo->actual_number_of_colors = numboxes; | |
638 | ||
639 | free(boxlist); //FIXME?? I don't know if this is correct - VS | |
640 | } | |
641 | ||
642 | ||
643 | /* | |
644 | * These routines are concerned with the time-critical task of mapping input | |
645 | * colors to the nearest color in the selected colormap. | |
646 | * | |
647 | * We re-use the histogram space as an "inverse color map", essentially a | |
648 | * cache for the results of nearest-color searches. All colors within a | |
649 | * histogram cell will be mapped to the same colormap entry, namely the one | |
650 | * closest to the cell's center. This may not be quite the closest entry to | |
651 | * the actual input color, but it's almost as good. A zero in the cache | |
652 | * indicates we haven't found the nearest color for that cell yet; the array | |
653 | * is cleared to zeroes before starting the mapping pass. When we find the | |
654 | * nearest color for a cell, its colormap index plus one is recorded in the | |
655 | * cache for future use. The pass2 scanning routines call fill_inverse_cmap | |
656 | * when they need to use an unfilled entry in the cache. | |
657 | * | |
658 | * Our method of efficiently finding nearest colors is based on the "locally | |
659 | * sorted search" idea described by Heckbert and on the incremental distance | |
660 | * calculation described by Spencer W. Thomas in chapter III.1 of Graphics | |
661 | * Gems II (James Arvo, ed. Academic Press, 1991). Thomas points out that | |
662 | * the distances from a given colormap entry to each cell of the histogram can | |
663 | * be computed quickly using an incremental method: the differences between | |
664 | * distances to adjacent cells themselves differ by a constant. This allows a | |
665 | * fairly fast implementation of the "brute force" approach of computing the | |
666 | * distance from every colormap entry to every histogram cell. Unfortunately, | |
667 | * it needs a work array to hold the best-distance-so-far for each histogram | |
668 | * cell (because the inner loop has to be over cells, not colormap entries). | |
669 | * The work array elements have to be INT32s, so the work array would need | |
670 | * 256Kb at our recommended precision. This is not feasible in DOS machines. | |
671 | * | |
672 | * To get around these problems, we apply Thomas' method to compute the | |
673 | * nearest colors for only the cells within a small subbox of the histogram. | |
674 | * The work array need be only as big as the subbox, so the memory usage | |
675 | * problem is solved. Furthermore, we need not fill subboxes that are never | |
676 | * referenced in pass2; many images use only part of the color gamut, so a | |
677 | * fair amount of work is saved. An additional advantage of this | |
678 | * approach is that we can apply Heckbert's locality criterion to quickly | |
679 | * eliminate colormap entries that are far away from the subbox; typically | |
680 | * three-fourths of the colormap entries are rejected by Heckbert's criterion, | |
681 | * and we need not compute their distances to individual cells in the subbox. | |
682 | * The speed of this approach is heavily influenced by the subbox size: too | |
683 | * small means too much overhead, too big loses because Heckbert's criterion | |
684 | * can't eliminate as many colormap entries. Empirically the best subbox | |
685 | * size seems to be about 1/512th of the histogram (1/8th in each direction). | |
686 | * | |
687 | * Thomas' article also describes a refined method which is asymptotically | |
688 | * faster than the brute-force method, but it is also far more complex and | |
689 | * cannot efficiently be applied to small subboxes. It is therefore not | |
690 | * useful for programs intended to be portable to DOS machines. On machines | |
691 | * with plenty of memory, filling the whole histogram in one shot with Thomas' | |
692 | * refined method might be faster than the present code --- but then again, | |
693 | * it might not be any faster, and it's certainly more complicated. | |
694 | */ | |
695 | ||
696 | ||
697 | /* log2(histogram cells in update box) for each axis; this can be adjusted */ | |
698 | #define BOX_C0_LOG (HIST_C0_BITS-3) | |
699 | #define BOX_C1_LOG (HIST_C1_BITS-3) | |
700 | #define BOX_C2_LOG (HIST_C2_BITS-3) | |
701 | ||
702 | #define BOX_C0_ELEMS (1<<BOX_C0_LOG) /* # of hist cells in update box */ | |
703 | #define BOX_C1_ELEMS (1<<BOX_C1_LOG) | |
704 | #define BOX_C2_ELEMS (1<<BOX_C2_LOG) | |
705 | ||
706 | #define BOX_C0_SHIFT (C0_SHIFT + BOX_C0_LOG) | |
707 | #define BOX_C1_SHIFT (C1_SHIFT + BOX_C1_LOG) | |
708 | #define BOX_C2_SHIFT (C2_SHIFT + BOX_C2_LOG) | |
709 | ||
710 | ||
711 | /* | |
712 | * The next three routines implement inverse colormap filling. They could | |
713 | * all be folded into one big routine, but splitting them up this way saves | |
714 | * some stack space (the mindist[] and bestdist[] arrays need not coexist) | |
715 | * and may allow some compilers to produce better code by registerizing more | |
716 | * inner-loop variables. | |
717 | */ | |
718 | ||
719 | static int | |
720 | find_nearby_colors (j_decompress_ptr cinfo, int minc0, int minc1, int minc2, | |
721 | JSAMPLE colorlist[]) | |
722 | /* Locate the colormap entries close enough to an update box to be candidates | |
723 | * for the nearest entry to some cell(s) in the update box. The update box | |
724 | * is specified by the center coordinates of its first cell. The number of | |
725 | * candidate colormap entries is returned, and their colormap indexes are | |
726 | * placed in colorlist[]. | |
727 | * This routine uses Heckbert's "locally sorted search" criterion to select | |
728 | * the colors that need further consideration. | |
729 | */ | |
730 | { | |
731 | int numcolors = cinfo->actual_number_of_colors; | |
732 | int maxc0, maxc1, maxc2; | |
733 | int centerc0, centerc1, centerc2; | |
734 | int i, x, ncolors; | |
735 | INT32 minmaxdist, min_dist, max_dist, tdist; | |
736 | INT32 mindist[MAXNUMCOLORS]; /* min distance to colormap entry i */ | |
737 | ||
738 | /* Compute true coordinates of update box's upper corner and center. | |
739 | * Actually we compute the coordinates of the center of the upper-corner | |
740 | * histogram cell, which are the upper bounds of the volume we care about. | |
741 | * Note that since ">>" rounds down, the "center" values may be closer to | |
742 | * min than to max; hence comparisons to them must be "<=", not "<". | |
743 | */ | |
744 | maxc0 = minc0 + ((1 << BOX_C0_SHIFT) - (1 << C0_SHIFT)); | |
745 | centerc0 = (minc0 + maxc0) >> 1; | |
746 | maxc1 = minc1 + ((1 << BOX_C1_SHIFT) - (1 << C1_SHIFT)); | |
747 | centerc1 = (minc1 + maxc1) >> 1; | |
748 | maxc2 = minc2 + ((1 << BOX_C2_SHIFT) - (1 << C2_SHIFT)); | |
749 | centerc2 = (minc2 + maxc2) >> 1; | |
750 | ||
751 | /* For each color in colormap, find: | |
752 | * 1. its minimum squared-distance to any point in the update box | |
753 | * (zero if color is within update box); | |
754 | * 2. its maximum squared-distance to any point in the update box. | |
755 | * Both of these can be found by considering only the corners of the box. | |
756 | * We save the minimum distance for each color in mindist[]; | |
757 | * only the smallest maximum distance is of interest. | |
758 | */ | |
759 | minmaxdist = 0x7FFFFFFFL; | |
760 | ||
761 | for (i = 0; i < numcolors; i++) { | |
762 | /* We compute the squared-c0-distance term, then add in the other two. */ | |
763 | x = GETJSAMPLE(cinfo->colormap[0][i]); | |
764 | if (x < minc0) { | |
765 | tdist = (x - minc0) * C0_SCALE; | |
766 | min_dist = tdist*tdist; | |
767 | tdist = (x - maxc0) * C0_SCALE; | |
768 | max_dist = tdist*tdist; | |
769 | } else if (x > maxc0) { | |
770 | tdist = (x - maxc0) * C0_SCALE; | |
771 | min_dist = tdist*tdist; | |
772 | tdist = (x - minc0) * C0_SCALE; | |
773 | max_dist = tdist*tdist; | |
774 | } else { | |
775 | /* within cell range so no contribution to min_dist */ | |
776 | min_dist = 0; | |
777 | if (x <= centerc0) { | |
778 | tdist = (x - maxc0) * C0_SCALE; | |
779 | max_dist = tdist*tdist; | |
780 | } else { | |
781 | tdist = (x - minc0) * C0_SCALE; | |
782 | max_dist = tdist*tdist; | |
783 | } | |
784 | } | |
785 | ||
786 | x = GETJSAMPLE(cinfo->colormap[1][i]); | |
787 | if (x < minc1) { | |
788 | tdist = (x - minc1) * C1_SCALE; | |
789 | min_dist += tdist*tdist; | |
790 | tdist = (x - maxc1) * C1_SCALE; | |
791 | max_dist += tdist*tdist; | |
792 | } else if (x > maxc1) { | |
793 | tdist = (x - maxc1) * C1_SCALE; | |
794 | min_dist += tdist*tdist; | |
795 | tdist = (x - minc1) * C1_SCALE; | |
796 | max_dist += tdist*tdist; | |
797 | } else { | |
798 | /* within cell range so no contribution to min_dist */ | |
799 | if (x <= centerc1) { | |
800 | tdist = (x - maxc1) * C1_SCALE; | |
801 | max_dist += tdist*tdist; | |
802 | } else { | |
803 | tdist = (x - minc1) * C1_SCALE; | |
804 | max_dist += tdist*tdist; | |
805 | } | |
806 | } | |
807 | ||
808 | x = GETJSAMPLE(cinfo->colormap[2][i]); | |
809 | if (x < minc2) { | |
810 | tdist = (x - minc2) * C2_SCALE; | |
811 | min_dist += tdist*tdist; | |
812 | tdist = (x - maxc2) * C2_SCALE; | |
813 | max_dist += tdist*tdist; | |
814 | } else if (x > maxc2) { | |
815 | tdist = (x - maxc2) * C2_SCALE; | |
816 | min_dist += tdist*tdist; | |
817 | tdist = (x - minc2) * C2_SCALE; | |
818 | max_dist += tdist*tdist; | |
819 | } else { | |
820 | /* within cell range so no contribution to min_dist */ | |
821 | if (x <= centerc2) { | |
822 | tdist = (x - maxc2) * C2_SCALE; | |
823 | max_dist += tdist*tdist; | |
824 | } else { | |
825 | tdist = (x - minc2) * C2_SCALE; | |
826 | max_dist += tdist*tdist; | |
827 | } | |
828 | } | |
829 | ||
830 | mindist[i] = min_dist; /* save away the results */ | |
831 | if (max_dist < minmaxdist) | |
832 | minmaxdist = max_dist; | |
833 | } | |
834 | ||
835 | /* Now we know that no cell in the update box is more than minmaxdist | |
836 | * away from some colormap entry. Therefore, only colors that are | |
837 | * within minmaxdist of some part of the box need be considered. | |
838 | */ | |
839 | ncolors = 0; | |
840 | for (i = 0; i < numcolors; i++) { | |
841 | if (mindist[i] <= minmaxdist) | |
842 | colorlist[ncolors++] = (JSAMPLE) i; | |
843 | } | |
844 | return ncolors; | |
845 | } | |
846 | ||
847 | ||
848 | static void | |
849 | find_best_colors (j_decompress_ptr cinfo, int minc0, int minc1, int minc2, | |
850 | int numcolors, JSAMPLE colorlist[], JSAMPLE bestcolor[]) | |
851 | /* Find the closest colormap entry for each cell in the update box, | |
852 | * given the list of candidate colors prepared by find_nearby_colors. | |
853 | * Return the indexes of the closest entries in the bestcolor[] array. | |
854 | * This routine uses Thomas' incremental distance calculation method to | |
855 | * find the distance from a colormap entry to successive cells in the box. | |
856 | */ | |
857 | { | |
858 | int ic0, ic1, ic2; | |
859 | int i, icolor; | |
860 | register INT32 * bptr; /* pointer into bestdist[] array */ | |
861 | JSAMPLE * cptr; /* pointer into bestcolor[] array */ | |
862 | INT32 dist0, dist1; /* initial distance values */ | |
863 | register INT32 dist2; /* current distance in inner loop */ | |
864 | INT32 xx0, xx1; /* distance increments */ | |
865 | register INT32 xx2; | |
866 | INT32 inc0, inc1, inc2; /* initial values for increments */ | |
867 | /* This array holds the distance to the nearest-so-far color for each cell */ | |
868 | INT32 bestdist[BOX_C0_ELEMS * BOX_C1_ELEMS * BOX_C2_ELEMS]; | |
869 | ||
870 | /* Initialize best-distance for each cell of the update box */ | |
871 | bptr = bestdist; | |
872 | for (i = BOX_C0_ELEMS*BOX_C1_ELEMS*BOX_C2_ELEMS-1; i >= 0; i--) | |
873 | *bptr++ = 0x7FFFFFFFL; | |
874 | ||
875 | /* For each color selected by find_nearby_colors, | |
876 | * compute its distance to the center of each cell in the box. | |
877 | * If that's less than best-so-far, update best distance and color number. | |
878 | */ | |
879 | ||
880 | /* Nominal steps between cell centers ("x" in Thomas article) */ | |
881 | #define STEP_C0 ((1 << C0_SHIFT) * C0_SCALE) | |
882 | #define STEP_C1 ((1 << C1_SHIFT) * C1_SCALE) | |
883 | #define STEP_C2 ((1 << C2_SHIFT) * C2_SCALE) | |
884 | ||
885 | for (i = 0; i < numcolors; i++) { | |
886 | icolor = GETJSAMPLE(colorlist[i]); | |
887 | /* Compute (square of) distance from minc0/c1/c2 to this color */ | |
888 | inc0 = (minc0 - GETJSAMPLE(cinfo->colormap[0][icolor])) * C0_SCALE; | |
889 | dist0 = inc0*inc0; | |
890 | inc1 = (minc1 - GETJSAMPLE(cinfo->colormap[1][icolor])) * C1_SCALE; | |
891 | dist0 += inc1*inc1; | |
892 | inc2 = (minc2 - GETJSAMPLE(cinfo->colormap[2][icolor])) * C2_SCALE; | |
893 | dist0 += inc2*inc2; | |
894 | /* Form the initial difference increments */ | |
895 | inc0 = inc0 * (2 * STEP_C0) + STEP_C0 * STEP_C0; | |
896 | inc1 = inc1 * (2 * STEP_C1) + STEP_C1 * STEP_C1; | |
897 | inc2 = inc2 * (2 * STEP_C2) + STEP_C2 * STEP_C2; | |
898 | /* Now loop over all cells in box, updating distance per Thomas method */ | |
899 | bptr = bestdist; | |
900 | cptr = bestcolor; | |
901 | xx0 = inc0; | |
902 | for (ic0 = BOX_C0_ELEMS-1; ic0 >= 0; ic0--) { | |
903 | dist1 = dist0; | |
904 | xx1 = inc1; | |
905 | for (ic1 = BOX_C1_ELEMS-1; ic1 >= 0; ic1--) { | |
906 | dist2 = dist1; | |
907 | xx2 = inc2; | |
908 | for (ic2 = BOX_C2_ELEMS-1; ic2 >= 0; ic2--) { | |
909 | if (dist2 < *bptr) { | |
910 | *bptr = dist2; | |
911 | *cptr = (JSAMPLE) icolor; | |
912 | } | |
913 | dist2 += xx2; | |
914 | xx2 += 2 * STEP_C2 * STEP_C2; | |
915 | bptr++; | |
916 | cptr++; | |
917 | } | |
918 | dist1 += xx1; | |
919 | xx1 += 2 * STEP_C1 * STEP_C1; | |
920 | } | |
921 | dist0 += xx0; | |
922 | xx0 += 2 * STEP_C0 * STEP_C0; | |
923 | } | |
924 | } | |
925 | } | |
926 | ||
927 | ||
928 | static void | |
929 | fill_inverse_cmap (j_decompress_ptr cinfo, int c0, int c1, int c2) | |
930 | /* Fill the inverse-colormap entries in the update box that contains */ | |
931 | /* histogram cell c0/c1/c2. (Only that one cell MUST be filled, but */ | |
932 | /* we can fill as many others as we wish.) */ | |
933 | { | |
934 | my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize; | |
935 | hist3d histogram = cquantize->histogram; | |
936 | int minc0, minc1, minc2; /* lower left corner of update box */ | |
937 | int ic0, ic1, ic2; | |
938 | register JSAMPLE * cptr; /* pointer into bestcolor[] array */ | |
939 | register histptr cachep; /* pointer into main cache array */ | |
940 | /* This array lists the candidate colormap indexes. */ | |
941 | JSAMPLE colorlist[MAXNUMCOLORS]; | |
942 | int numcolors; /* number of candidate colors */ | |
943 | /* This array holds the actually closest colormap index for each cell. */ | |
944 | JSAMPLE bestcolor[BOX_C0_ELEMS * BOX_C1_ELEMS * BOX_C2_ELEMS]; | |
945 | ||
946 | /* Convert cell coordinates to update box ID */ | |
947 | c0 >>= BOX_C0_LOG; | |
948 | c1 >>= BOX_C1_LOG; | |
949 | c2 >>= BOX_C2_LOG; | |
950 | ||
951 | /* Compute true coordinates of update box's origin corner. | |
952 | * Actually we compute the coordinates of the center of the corner | |
953 | * histogram cell, which are the lower bounds of the volume we care about. | |
954 | */ | |
955 | minc0 = (c0 << BOX_C0_SHIFT) + ((1 << C0_SHIFT) >> 1); | |
956 | minc1 = (c1 << BOX_C1_SHIFT) + ((1 << C1_SHIFT) >> 1); | |
957 | minc2 = (c2 << BOX_C2_SHIFT) + ((1 << C2_SHIFT) >> 1); | |
958 | ||
959 | /* Determine which colormap entries are close enough to be candidates | |
960 | * for the nearest entry to some cell in the update box. | |
961 | */ | |
962 | numcolors = find_nearby_colors(cinfo, minc0, minc1, minc2, colorlist); | |
963 | ||
964 | /* Determine the actually nearest colors. */ | |
965 | find_best_colors(cinfo, minc0, minc1, minc2, numcolors, colorlist, | |
966 | bestcolor); | |
967 | ||
968 | /* Save the best color numbers (plus 1) in the main cache array */ | |
969 | c0 <<= BOX_C0_LOG; /* convert ID back to base cell indexes */ | |
970 | c1 <<= BOX_C1_LOG; | |
971 | c2 <<= BOX_C2_LOG; | |
972 | cptr = bestcolor; | |
973 | for (ic0 = 0; ic0 < BOX_C0_ELEMS; ic0++) { | |
974 | for (ic1 = 0; ic1 < BOX_C1_ELEMS; ic1++) { | |
975 | cachep = & histogram[c0+ic0][c1+ic1][c2]; | |
976 | for (ic2 = 0; ic2 < BOX_C2_ELEMS; ic2++) { | |
977 | *cachep++ = (histcell) (GETJSAMPLE(*cptr++) + 1); | |
978 | } | |
979 | } | |
980 | } | |
981 | } | |
982 | ||
983 | ||
984 | /* | |
985 | * Map some rows of pixels to the output colormapped representation. | |
986 | */ | |
987 | ||
988 | void | |
989 | pass2_no_dither (j_decompress_ptr cinfo, | |
990 | JSAMPARRAY input_buf, JSAMPARRAY output_buf, int num_rows) | |
991 | /* This version performs no dithering */ | |
992 | { | |
993 | my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize; | |
994 | hist3d histogram = cquantize->histogram; | |
995 | register JSAMPROW inptr, outptr; | |
996 | register histptr cachep; | |
997 | register int c0, c1, c2; | |
998 | int row; | |
999 | JDIMENSION col; | |
1000 | JDIMENSION width = cinfo->output_width; | |
1001 | ||
1002 | for (row = 0; row < num_rows; row++) { | |
1003 | inptr = input_buf[row]; | |
1004 | outptr = output_buf[row]; | |
1005 | for (col = width; col > 0; col--) { | |
1006 | /* get pixel value and index into the cache */ | |
1007 | c0 = GETJSAMPLE(*inptr++) >> C0_SHIFT; | |
1008 | c1 = GETJSAMPLE(*inptr++) >> C1_SHIFT; | |
1009 | c2 = GETJSAMPLE(*inptr++) >> C2_SHIFT; | |
1010 | cachep = & histogram[c0][c1][c2]; | |
1011 | /* If we have not seen this color before, find nearest colormap entry */ | |
1012 | /* and update the cache */ | |
1013 | if (*cachep == 0) | |
1014 | fill_inverse_cmap(cinfo, c0,c1,c2); | |
1015 | /* Now emit the colormap index for this cell */ | |
1016 | *outptr++ = (JSAMPLE) (*cachep - 1); | |
1017 | } | |
1018 | } | |
1019 | } | |
1020 | ||
1021 | ||
1022 | void | |
1023 | pass2_fs_dither (j_decompress_ptr cinfo, | |
1024 | JSAMPARRAY input_buf, JSAMPARRAY output_buf, int num_rows) | |
1025 | /* This version performs Floyd-Steinberg dithering */ | |
1026 | { | |
1027 | my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize; | |
1028 | hist3d histogram = cquantize->histogram; | |
1029 | register LOCFSERROR cur0, cur1, cur2; /* current error or pixel value */ | |
1030 | LOCFSERROR belowerr0, belowerr1, belowerr2; /* error for pixel below cur */ | |
1031 | LOCFSERROR bpreverr0, bpreverr1, bpreverr2; /* error for below/prev col */ | |
1032 | register FSERRPTR errorptr; /* => fserrors[] at column before current */ | |
1033 | JSAMPROW inptr; /* => current input pixel */ | |
1034 | JSAMPROW outptr; /* => current output pixel */ | |
1035 | histptr cachep; | |
1036 | int dir; /* +1 or -1 depending on direction */ | |
1037 | int dir3; /* 3*dir, for advancing inptr & errorptr */ | |
1038 | int row; | |
1039 | JDIMENSION col; | |
1040 | JDIMENSION width = cinfo->output_width; | |
1041 | JSAMPLE *range_limit = cinfo->sample_range_limit; | |
1042 | int *error_limit = cquantize->error_limiter; | |
1043 | JSAMPROW colormap0 = cinfo->colormap[0]; | |
1044 | JSAMPROW colormap1 = cinfo->colormap[1]; | |
1045 | JSAMPROW colormap2 = cinfo->colormap[2]; | |
1046 | ||
1047 | ||
1048 | for (row = 0; row < num_rows; row++) { | |
1049 | inptr = input_buf[row]; | |
1050 | outptr = output_buf[row]; | |
1051 | if (cquantize->on_odd_row) { | |
1052 | /* work right to left in this row */ | |
1053 | inptr += (width-1) * 3; /* so point to rightmost pixel */ | |
1054 | outptr += width-1; | |
1055 | dir = -1; | |
1056 | dir3 = -3; | |
1057 | errorptr = cquantize->fserrors + (width+1)*3; /* => entry after last column */ | |
1058 | cquantize->on_odd_row = false; /* flip for next time */ | |
1059 | } else { | |
1060 | /* work left to right in this row */ | |
1061 | dir = 1; | |
1062 | dir3 = 3; | |
1063 | errorptr = cquantize->fserrors; /* => entry before first real column */ | |
1064 | cquantize->on_odd_row = true; /* flip for next time */ | |
1065 | } | |
1066 | /* Preset error values: no error propagated to first pixel from left */ | |
1067 | cur0 = cur1 = cur2 = 0; | |
1068 | /* and no error propagated to row below yet */ | |
1069 | belowerr0 = belowerr1 = belowerr2 = 0; | |
1070 | bpreverr0 = bpreverr1 = bpreverr2 = 0; | |
1071 | ||
1072 | for (col = width; col > 0; col--) { | |
1073 | /* curN holds the error propagated from the previous pixel on the | |
1074 | * current line. Add the error propagated from the previous line | |
1075 | * to form the complete error correction term for this pixel, and | |
1076 | * round the error term (which is expressed * 16) to an integer. | |
1077 | * RIGHT_SHIFT rounds towards minus infinity, so adding 8 is correct | |
1078 | * for either sign of the error value. | |
1079 | * Note: errorptr points to *previous* column's array entry. | |
1080 | */ | |
1081 | cur0 = RIGHT_SHIFT(cur0 + errorptr[dir3+0] + 8, 4); | |
1082 | cur1 = RIGHT_SHIFT(cur1 + errorptr[dir3+1] + 8, 4); | |
1083 | cur2 = RIGHT_SHIFT(cur2 + errorptr[dir3+2] + 8, 4); | |
1084 | /* Limit the error using transfer function set by init_error_limit. | |
1085 | * See comments with init_error_limit for rationale. | |
1086 | */ | |
1087 | cur0 = error_limit[cur0]; | |
1088 | cur1 = error_limit[cur1]; | |
1089 | cur2 = error_limit[cur2]; | |
1090 | /* Form pixel value + error, and range-limit to 0..MAXJSAMPLE. | |
1091 | * The maximum error is +- MAXJSAMPLE (or less with error limiting); | |
1092 | * this sets the required size of the range_limit array. | |
1093 | */ | |
1094 | cur0 += GETJSAMPLE(inptr[0]); | |
1095 | cur1 += GETJSAMPLE(inptr[1]); | |
1096 | cur2 += GETJSAMPLE(inptr[2]); | |
1097 | cur0 = GETJSAMPLE(range_limit[cur0]); | |
1098 | cur1 = GETJSAMPLE(range_limit[cur1]); | |
1099 | cur2 = GETJSAMPLE(range_limit[cur2]); | |
1100 | /* Index into the cache with adjusted pixel value */ | |
1101 | cachep = & histogram[cur0>>C0_SHIFT][cur1>>C1_SHIFT][cur2>>C2_SHIFT]; | |
1102 | /* If we have not seen this color before, find nearest colormap */ | |
1103 | /* entry and update the cache */ | |
1104 | if (*cachep == 0) | |
1105 | fill_inverse_cmap(cinfo, cur0>>C0_SHIFT,cur1>>C1_SHIFT,cur2>>C2_SHIFT); | |
1106 | /* Now emit the colormap index for this cell */ | |
1107 | { register int pixcode = *cachep - 1; | |
1108 | *outptr = (JSAMPLE) pixcode; | |
1109 | /* Compute representation error for this pixel */ | |
1110 | cur0 -= GETJSAMPLE(colormap0[pixcode]); | |
1111 | cur1 -= GETJSAMPLE(colormap1[pixcode]); | |
1112 | cur2 -= GETJSAMPLE(colormap2[pixcode]); | |
1113 | } | |
1114 | /* Compute error fractions to be propagated to adjacent pixels. | |
1115 | * Add these into the running sums, and simultaneously shift the | |
1116 | * next-line error sums left by 1 column. | |
1117 | */ | |
1118 | { register LOCFSERROR bnexterr, delta; | |
1119 | ||
1120 | bnexterr = cur0; /* Process component 0 */ | |
1121 | delta = cur0 * 2; | |
1122 | cur0 += delta; /* form error * 3 */ | |
1123 | errorptr[0] = (FSERROR) (bpreverr0 + cur0); | |
1124 | cur0 += delta; /* form error * 5 */ | |
1125 | bpreverr0 = belowerr0 + cur0; | |
1126 | belowerr0 = bnexterr; | |
1127 | cur0 += delta; /* form error * 7 */ | |
1128 | bnexterr = cur1; /* Process component 1 */ | |
1129 | delta = cur1 * 2; | |
1130 | cur1 += delta; /* form error * 3 */ | |
1131 | errorptr[1] = (FSERROR) (bpreverr1 + cur1); | |
1132 | cur1 += delta; /* form error * 5 */ | |
1133 | bpreverr1 = belowerr1 + cur1; | |
1134 | belowerr1 = bnexterr; | |
1135 | cur1 += delta; /* form error * 7 */ | |
1136 | bnexterr = cur2; /* Process component 2 */ | |
1137 | delta = cur2 * 2; | |
1138 | cur2 += delta; /* form error * 3 */ | |
1139 | errorptr[2] = (FSERROR) (bpreverr2 + cur2); | |
1140 | cur2 += delta; /* form error * 5 */ | |
1141 | bpreverr2 = belowerr2 + cur2; | |
1142 | belowerr2 = bnexterr; | |
1143 | cur2 += delta; /* form error * 7 */ | |
1144 | } | |
1145 | /* At this point curN contains the 7/16 error value to be propagated | |
1146 | * to the next pixel on the current line, and all the errors for the | |
1147 | * next line have been shifted over. We are therefore ready to move on. | |
1148 | */ | |
1149 | inptr += dir3; /* Advance pixel pointers to next column */ | |
1150 | outptr += dir; | |
1151 | errorptr += dir3; /* advance errorptr to current column */ | |
1152 | } | |
1153 | /* Post-loop cleanup: we must unload the final error values into the | |
1154 | * final fserrors[] entry. Note we need not unload belowerrN because | |
1155 | * it is for the dummy column before or after the actual array. | |
1156 | */ | |
1157 | errorptr[0] = (FSERROR) bpreverr0; /* unload prev errs into array */ | |
1158 | errorptr[1] = (FSERROR) bpreverr1; | |
1159 | errorptr[2] = (FSERROR) bpreverr2; | |
1160 | } | |
1161 | } | |
1162 | ||
1163 | ||
1164 | /* | |
1165 | * Initialize the error-limiting transfer function (lookup table). | |
1166 | * The raw F-S error computation can potentially compute error values of up to | |
1167 | * +- MAXJSAMPLE. But we want the maximum correction applied to a pixel to be | |
1168 | * much less, otherwise obviously wrong pixels will be created. (Typical | |
1169 | * effects include weird fringes at color-area boundaries, isolated bright | |
1170 | * pixels in a dark area, etc.) The standard advice for avoiding this problem | |
1171 | * is to ensure that the "corners" of the color cube are allocated as output | |
1172 | * colors; then repeated errors in the same direction cannot cause cascading | |
1173 | * error buildup. However, that only prevents the error from getting | |
1174 | * completely out of hand; Aaron Giles reports that error limiting improves | |
1175 | * the results even with corner colors allocated. | |
1176 | * A simple clamping of the error values to about +- MAXJSAMPLE/8 works pretty | |
1177 | * well, but the smoother transfer function used below is even better. Thanks | |
1178 | * to Aaron Giles for this idea. | |
1179 | */ | |
1180 | ||
1181 | static void | |
1182 | init_error_limit (j_decompress_ptr cinfo) | |
1183 | /* Allocate and fill in the error_limiter table */ | |
1184 | { | |
1185 | my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize; | |
1186 | int * table; | |
1187 | int in, out; | |
1188 | ||
1189 | table = (int *) malloc((MAXJSAMPLE*2+1) * sizeof(int)); | |
1190 | table += MAXJSAMPLE; /* so can index -MAXJSAMPLE .. +MAXJSAMPLE */ | |
1191 | cquantize->error_limiter = table; | |
1192 | ||
1193 | #define STEPSIZE ((MAXJSAMPLE+1)/16) | |
1194 | /* Map errors 1:1 up to +- MAXJSAMPLE/16 */ | |
1195 | out = 0; | |
1196 | for (in = 0; in < STEPSIZE; in++, out++) { | |
1197 | table[in] = out; table[-in] = -out; | |
1198 | } | |
1199 | /* Map errors 1:2 up to +- 3*MAXJSAMPLE/16 */ | |
1200 | for (; in < STEPSIZE*3; in++, out += (in&1) ? 0 : 1) { | |
1201 | table[in] = out; table[-in] = -out; | |
1202 | } | |
1203 | /* Clamp the rest to final out value (which is (MAXJSAMPLE+1)/8) */ | |
1204 | for (; in <= MAXJSAMPLE; in++) { | |
1205 | table[in] = out; table[-in] = -out; | |
1206 | } | |
1207 | #undef STEPSIZE | |
1208 | } | |
1209 | ||
1210 | ||
1211 | /* | |
1212 | * Finish up at the end of each pass. | |
1213 | */ | |
1214 | ||
1215 | void | |
1216 | finish_pass1 (j_decompress_ptr cinfo) | |
1217 | { | |
1218 | my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize; | |
1219 | ||
1220 | /* Select the representative colors and fill in cinfo->colormap */ | |
1221 | cinfo->colormap = cquantize->sv_colormap; | |
1222 | select_colors(cinfo, cquantize->desired); | |
1223 | /* Force next pass to zero the color index table */ | |
1224 | cquantize->needs_zeroed = true; | |
1225 | } | |
1226 | ||
1227 | ||
1228 | void | |
1229 | finish_pass2 (j_decompress_ptr cinfo) | |
1230 | { | |
1231 | /* no work */ | |
1232 | } | |
1233 | ||
1234 | ||
1235 | /* | |
1236 | * Initialize for each processing pass. | |
1237 | */ | |
1238 | ||
1239 | void | |
1240 | start_pass_2_quant (j_decompress_ptr cinfo, bool is_pre_scan) | |
1241 | { | |
1242 | my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize; | |
1243 | hist3d histogram = cquantize->histogram; | |
1244 | int i; | |
1245 | ||
1246 | if (is_pre_scan) { | |
1247 | /* Set up method pointers */ | |
1248 | cquantize->pub.color_quantize = prescan_quantize; | |
1249 | cquantize->pub.finish_pass = finish_pass1; | |
1250 | cquantize->needs_zeroed = true; /* Always zero histogram */ | |
1251 | } else { | |
1252 | /* Set up method pointers */ | |
1253 | cquantize->pub.color_quantize = pass2_fs_dither; | |
1254 | cquantize->pub.finish_pass = finish_pass2; | |
1255 | ||
1256 | /* Make sure color count is acceptable */ | |
1257 | i = cinfo->actual_number_of_colors; | |
1258 | ||
1259 | { | |
1260 | size_t arraysize = (size_t) ((cinfo->output_width + 2) * | |
1261 | (3 * sizeof(FSERROR))); | |
1262 | /* Allocate Floyd-Steinberg workspace if we didn't already. */ | |
1263 | if (cquantize->fserrors == NULL) | |
1264 | cquantize->fserrors = (INT16*) malloc(arraysize); | |
1265 | /* Initialize the propagated errors to zero. */ | |
1266 | memset((void *) cquantize->fserrors, 0, arraysize); | |
1267 | /* Make the error-limit table if we didn't already. */ | |
1268 | if (cquantize->error_limiter == NULL) | |
1269 | init_error_limit(cinfo); | |
1270 | cquantize->on_odd_row = false; | |
1271 | } | |
1272 | ||
1273 | } | |
1274 | /* Zero the histogram or inverse color map, if necessary */ | |
1275 | if (cquantize->needs_zeroed) { | |
1276 | for (i = 0; i < HIST_C0_ELEMS; i++) { | |
1277 | memset((void *) histogram[i], 0, | |
1278 | HIST_C1_ELEMS*HIST_C2_ELEMS * sizeof(histcell)); | |
1279 | } | |
1280 | cquantize->needs_zeroed = false; | |
1281 | } | |
1282 | } | |
1283 | ||
1284 | ||
1285 | /* | |
1286 | * Switch to a new external colormap between output passes. | |
1287 | */ | |
1288 | ||
1289 | void | |
1290 | new_color_map_2_quant (j_decompress_ptr cinfo) | |
1291 | { | |
1292 | my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize; | |
1293 | ||
1294 | /* Reset the inverse color map */ | |
1295 | cquantize->needs_zeroed = true; | |
1296 | } | |
1297 | ||
1298 | ||
1299 | /* | |
1300 | * Module initialization routine for 2-pass color quantization. | |
1301 | */ | |
1302 | ||
1303 | void | |
1304 | jinit_2pass_quantizer (j_decompress_ptr cinfo) | |
1305 | { | |
1306 | my_cquantize_ptr cquantize; | |
1307 | int i; | |
1308 | ||
1309 | cquantize = (my_cquantize_ptr) malloc(sizeof(my_cquantizer)); | |
1310 | cinfo->cquantize = (struct jpeg_color_quantizer *) cquantize; | |
1311 | cquantize->pub.start_pass = start_pass_2_quant; | |
1312 | cquantize->pub.new_color_map = new_color_map_2_quant; | |
1313 | cquantize->fserrors = NULL; /* flag optional arrays not allocated */ | |
1314 | cquantize->error_limiter = NULL; | |
1315 | ||
1316 | ||
1317 | /* Allocate the histogram/inverse colormap storage */ | |
1318 | cquantize->histogram = (hist3d) malloc(HIST_C0_ELEMS * sizeof(hist2d)); | |
1319 | for (i = 0; i < HIST_C0_ELEMS; i++) { | |
1320 | cquantize->histogram[i] = (hist2d) malloc(HIST_C1_ELEMS*HIST_C2_ELEMS * sizeof(histcell)); | |
1321 | } | |
1322 | cquantize->needs_zeroed = true; /* histogram is garbage now */ | |
1323 | ||
1324 | /* Allocate storage for the completed colormap, if required. | |
1325 | * We do this now since it is storage and may affect | |
1326 | * the memory manager's space calculations. | |
1327 | */ | |
1328 | { | |
1329 | /* Make sure color count is acceptable */ | |
1330 | int desired = cinfo->desired_number_of_colors; | |
1331 | ||
1332 | cquantize->sv_colormap = (JSAMPARRAY) malloc(sizeof(JSAMPROW) * 3); | |
1333 | cquantize->sv_colormap[0] = (JSAMPROW) malloc(sizeof(JSAMPLE) * desired); | |
1334 | cquantize->sv_colormap[1] = (JSAMPROW) malloc(sizeof(JSAMPLE) * desired); | |
1335 | cquantize->sv_colormap[2] = (JSAMPROW) malloc(sizeof(JSAMPLE) * desired); | |
1336 | ||
1337 | cquantize->desired = desired; | |
1338 | } | |
1339 | ||
1340 | /* Allocate Floyd-Steinberg workspace if necessary. | |
1341 | * This isn't really needed until pass 2, but again it is storage. | |
1342 | * Although we will cope with a later change in dither_mode, | |
1343 | * we do not promise to honor max_memory_to_use if dither_mode changes. | |
1344 | */ | |
1345 | { | |
1346 | cquantize->fserrors = (FSERRPTR) malloc( | |
1347 | (size_t) ((cinfo->output_width + 2) * (3 * sizeof(FSERROR)))); | |
1348 | /* Might as well create the error-limiting table too. */ | |
1349 | init_error_limit(cinfo); | |
1350 | } | |
1351 | } | |
1352 | ||
1353 | ||
1354 | ||
1355 | ||
1356 | ||
1357 | ||
1358 | ||
1359 | ||
1360 | ||
1361 | ||
1362 | void | |
1363 | prepare_range_limit_table (j_decompress_ptr cinfo) | |
1364 | /* Allocate and fill in the sample_range_limit table */ | |
1365 | { | |
1366 | JSAMPLE * table; | |
1367 | int i; | |
1368 | ||
1369 | table = (JSAMPLE *) malloc((5 * (MAXJSAMPLE+1) + CENTERJSAMPLE) * sizeof(JSAMPLE)); | |
1370 | cinfo->srl_orig = table; | |
1371 | table += (MAXJSAMPLE+1); /* allow negative subscripts of simple table */ | |
1372 | cinfo->sample_range_limit = table; | |
1373 | /* First segment of "simple" table: limit[x] = 0 for x < 0 */ | |
1374 | memset(table - (MAXJSAMPLE+1), 0, (MAXJSAMPLE+1) * sizeof(JSAMPLE)); | |
1375 | /* Main part of "simple" table: limit[x] = x */ | |
1376 | for (i = 0; i <= MAXJSAMPLE; i++) | |
1377 | table[i] = (JSAMPLE) i; | |
1378 | table += CENTERJSAMPLE; /* Point to where post-IDCT table starts */ | |
1379 | /* End of simple table, rest of first half of post-IDCT table */ | |
1380 | for (i = CENTERJSAMPLE; i < 2*(MAXJSAMPLE+1); i++) | |
1381 | table[i] = MAXJSAMPLE; | |
1382 | /* Second half of post-IDCT table */ | |
1383 | memset(table + (2 * (MAXJSAMPLE+1)), 0, | |
1384 | (2 * (MAXJSAMPLE+1) - CENTERJSAMPLE) * sizeof(JSAMPLE)); | |
1385 | memcpy(table + (4 * (MAXJSAMPLE+1) - CENTERJSAMPLE), | |
1386 | cinfo->sample_range_limit, CENTERJSAMPLE * sizeof(JSAMPLE)); | |
1387 | } | |
1388 | ||
1389 | ||
1390 | ||
1391 | ||
1392 | /* | |
1393 | * wxQuantize | |
1394 | */ | |
1395 | ||
1396 | IMPLEMENT_DYNAMIC_CLASS(wxQuantize, wxObject) | |
1397 | ||
1398 | void wxQuantize::DoQuantize(unsigned w, unsigned h, unsigned char **in_rows, unsigned char **out_rows, | |
1399 | unsigned char *palette, int desiredNoColours) | |
1400 | { | |
1401 | j_decompress dec; | |
1402 | my_cquantize_ptr cquantize; | |
1403 | ||
1404 | dec.output_width = w; | |
1405 | dec.desired_number_of_colors = desiredNoColours; | |
1406 | prepare_range_limit_table(&dec); | |
1407 | jinit_2pass_quantizer(&dec); | |
1408 | cquantize = (my_cquantize_ptr) dec.cquantize; | |
1409 | ||
1410 | ||
1411 | cquantize->pub.start_pass(&dec, true); | |
1412 | cquantize->pub.color_quantize(&dec, in_rows, out_rows, h); | |
1413 | cquantize->pub.finish_pass(&dec); | |
1414 | ||
1415 | cquantize->pub.start_pass(&dec, false); | |
1416 | cquantize->pub.color_quantize(&dec, in_rows, out_rows, h); | |
1417 | cquantize->pub.finish_pass(&dec); | |
1418 | ||
1419 | ||
1420 | for (int i = 0; i < dec.desired_number_of_colors; i++) { | |
1421 | palette[3 * i + 0] = dec.colormap[0][i]; | |
1422 | palette[3 * i + 1] = dec.colormap[1][i]; | |
1423 | palette[3 * i + 2] = dec.colormap[2][i]; | |
1424 | } | |
1425 | ||
1426 | for (int ii = 0; ii < HIST_C0_ELEMS; ii++) free(cquantize->histogram[ii]); | |
1427 | free(cquantize->histogram); | |
1428 | free(dec.colormap[0]); | |
1429 | free(dec.colormap[1]); | |
1430 | free(dec.colormap[2]); | |
1431 | free(dec.colormap); | |
1432 | free(dec.srl_orig); | |
1433 | ||
1434 | //free(cquantize->error_limiter); | |
1435 | free((void*)(cquantize->error_limiter - MAXJSAMPLE)); // To reverse what was done to it | |
1436 | ||
1437 | free(cquantize->fserrors); | |
1438 | free(cquantize); | |
1439 | } | |
1440 | ||
1441 | // TODO: somehow make use of the Windows system colours, rather than ignoring them for the | |
1442 | // purposes of quantization. | |
1443 | ||
1444 | bool wxQuantize::Quantize(const wxImage& src, wxImage& dest, wxPalette** pPalette, int desiredNoColours, | |
1445 | unsigned char** eightBitData, int flags) | |
1446 | ||
1447 | { | |
1448 | int i; | |
1449 | int w = src.GetWidth(); | |
1450 | int h = src.GetHeight(); | |
1451 | ||
1452 | int windowsSystemColourCount = 20; | |
1453 | int paletteShift = 0; | |
1454 | ||
1455 | // Shift the palette up by the number of Windows system colours, | |
1456 | // if necessary | |
1457 | if (flags & wxQUANTIZE_INCLUDE_WINDOWS_COLOURS) | |
1458 | paletteShift = windowsSystemColourCount; | |
1459 | ||
1460 | // Make room for the Windows system colours | |
1461 | #ifdef __WXMSW__ | |
1462 | if ((flags & wxQUANTIZE_INCLUDE_WINDOWS_COLOURS) && (desiredNoColours > (256 - windowsSystemColourCount))) | |
1463 | desiredNoColours = 256 - windowsSystemColourCount; | |
1464 | #endif | |
1465 | ||
1466 | // create rows info: | |
1467 | unsigned char **rows = new unsigned char *[h]; | |
1468 | h = src.GetHeight(), w = src.GetWidth(); | |
1469 | unsigned char *imgdt = src.GetData(); | |
1470 | for (i = 0; i < h; i++) | |
1471 | rows[i] = imgdt + 3/*RGB*/ * w * i; | |
1472 | ||
1473 | unsigned char palette[3*256]; | |
1474 | ||
1475 | // This is the image as represented by palette indexes. | |
1476 | unsigned char *data8bit = new unsigned char[w * h]; | |
1477 | unsigned char **outrows = new unsigned char *[h]; | |
1478 | for (i = 0; i < h; i++) | |
1479 | outrows[i] = data8bit + w * i; | |
1480 | ||
1481 | //RGB->palette | |
1482 | DoQuantize(w, h, rows, outrows, palette, desiredNoColours); | |
1483 | ||
1484 | delete[] rows; | |
1485 | delete[] outrows; | |
1486 | ||
1487 | // palette->RGB(max.256) | |
1488 | ||
1489 | if (flags & wxQUANTIZE_FILL_DESTINATION_IMAGE) | |
1490 | { | |
1491 | if (!dest.Ok()) | |
1492 | dest.Create(w, h); | |
1493 | ||
1494 | imgdt = dest.GetData(); | |
1495 | for (i = 0; i < w * h; i++) | |
1496 | { | |
1497 | unsigned char c = data8bit[i]; | |
1498 | imgdt[3 * i + 0/*R*/] = palette[3 * c + 0]; | |
1499 | imgdt[3 * i + 1/*G*/] = palette[3 * c + 1]; | |
1500 | imgdt[3 * i + 2/*B*/] = palette[3 * c + 2]; | |
1501 | } | |
1502 | } | |
1503 | ||
1504 | if (eightBitData && (flags & wxQUANTIZE_RETURN_8BIT_DATA)) | |
1505 | { | |
1506 | #ifdef __WXMSW__ | |
1507 | if (flags & wxQUANTIZE_INCLUDE_WINDOWS_COLOURS) | |
1508 | { | |
1509 | // We need to shift the palette entries up | |
1510 | // to make room for the Windows system colours. | |
1511 | for (i = 0; i < w * h; i++) | |
1512 | data8bit[i] = data8bit[i] + paletteShift; | |
1513 | } | |
1514 | #endif | |
1515 | *eightBitData = data8bit; | |
1516 | } | |
1517 | else | |
1518 | delete[] data8bit; | |
1519 | ||
1520 | // Make a wxWindows palette | |
1521 | if (pPalette) | |
1522 | { | |
1523 | unsigned char* r = new unsigned char[256]; | |
1524 | unsigned char* g = new unsigned char[256]; | |
1525 | unsigned char* b = new unsigned char[256]; | |
1526 | ||
1527 | #ifdef __WXMSW__ | |
1528 | // Fill the first 20 entries with Windows system colours | |
1529 | if (flags & wxQUANTIZE_INCLUDE_WINDOWS_COLOURS) | |
1530 | { | |
1531 | HDC hDC = ::GetDC(NULL); | |
1532 | PALETTEENTRY* entries = new PALETTEENTRY[windowsSystemColourCount]; | |
1533 | ::GetSystemPaletteEntries(hDC, 0, windowsSystemColourCount, entries); | |
1534 | ::ReleaseDC(NULL, hDC); | |
1535 | ||
1536 | for (i = 0; i < windowsSystemColourCount; i++) | |
1537 | { | |
1538 | r[i] = entries[i].peRed; | |
1539 | g[i] = entries[i].peGreen; | |
1540 | b[i] = entries[i].peBlue; | |
1541 | } | |
1542 | delete[] entries; | |
1543 | } | |
1544 | #endif | |
1545 | ||
1546 | for (i = 0; i < desiredNoColours; i++) | |
1547 | { | |
1548 | r[i+paletteShift] = palette[i*3 + 0]; | |
1549 | g[i+paletteShift] = palette[i*3 + 1]; | |
1550 | b[i+paletteShift] = palette[i*3 + 2]; | |
1551 | } | |
1552 | ||
1553 | // Blank out any remaining palette entries | |
1554 | for (i = desiredNoColours+paletteShift; i < 256; i++) | |
1555 | { | |
1556 | r[i] = 0; | |
1557 | g[i] = 0; | |
1558 | b[i] = 0; | |
1559 | } | |
1560 | *pPalette = new wxPalette(256, r, g, b); | |
1561 | delete[] r; | |
1562 | delete[] g; | |
1563 | delete[] b; | |
1564 | } | |
1565 | ||
1566 | return TRUE; | |
1567 | } | |
1568 | ||
1569 | // This version sets a palette in the destination image so you don't | |
1570 | // have to manage it yourself. | |
1571 | ||
1572 | bool wxQuantize::Quantize(const wxImage& src, wxImage& dest, int desiredNoColours, | |
1573 | unsigned char** eightBitData, int flags) | |
1574 | { | |
1575 | wxPalette* palette = NULL; | |
1576 | if (Quantize(src, dest, & palette, desiredNoColours, eightBitData, flags)) | |
1577 | { | |
1578 | if (palette) | |
1579 | { | |
1580 | dest.SetPalette(* palette); | |
1581 | delete palette; | |
1582 | } | |
1583 | return TRUE; | |
1584 | } | |
1585 | else | |
1586 | return FALSE; | |
1587 | } | |
1588 |