1 | /*
|
---|
2 |
|
---|
3 | gbmhist.c - Histogram/Frequency-of-use method of colour reduction
|
---|
4 |
|
---|
5 | */
|
---|
6 |
|
---|
7 | /*...sincludes:0:*/
|
---|
8 | #include <stdio.h>
|
---|
9 | #include <stddef.h>
|
---|
10 | #include <stdlib.h>
|
---|
11 | #include <string.h>
|
---|
12 | #include "gbm.h"
|
---|
13 |
|
---|
14 | /*...vgbm\46\h:0:*/
|
---|
15 | /*...e*/
|
---|
16 |
|
---|
17 | #define N_COLS 2049
|
---|
18 | #define N_HASH 5191
|
---|
19 | #define HASH(r,g,b) (word) ( (((r)+(g))*((g)+(b))*((b)+(r))) % N_HASH )
|
---|
20 |
|
---|
21 | typedef struct { byte b, g, r; dword freq; byte nearest; } FREQ;
|
---|
22 |
|
---|
23 | typedef struct
|
---|
24 | {
|
---|
25 | int n_cols;
|
---|
26 | byte rm, gm, bm;
|
---|
27 | FREQ f[N_COLS];
|
---|
28 | word ht[N_HASH];
|
---|
29 | } GBMHIST;
|
---|
30 |
|
---|
31 | /*...sgbm_create_hist \45\ create empty hist:0:*/
|
---|
32 | GBMHIST *gbm_create_hist(
|
---|
33 | byte rm, byte gm, byte bm
|
---|
34 | )
|
---|
35 | {
|
---|
36 | GBMHIST *hist;
|
---|
37 |
|
---|
38 | if ( (hist = malloc((size_t) sizeof(GBMHIST))) == NULL )
|
---|
39 | return NULL;
|
---|
40 | hist->rm = rm;
|
---|
41 | hist->gm = gm;
|
---|
42 | hist->bm = bm;
|
---|
43 | hist->n_cols = 0;
|
---|
44 | memset(hist->ht, 0xff, N_HASH * sizeof(word));
|
---|
45 | return hist;
|
---|
46 | }
|
---|
47 | /*...e*/
|
---|
48 | /*...sgbm_delete_hist \45\ delete hist:0:*/
|
---|
49 | void gbm_delete_hist(GBMHIST *hist)
|
---|
50 | {
|
---|
51 | free(hist);
|
---|
52 | }
|
---|
53 | /*...e*/
|
---|
54 | /*...sgbm_add_to_hist \45\ add bitmap data to hist:0:*/
|
---|
55 | BOOLEAN gbm_add_to_hist(
|
---|
56 | GBMHIST *hist,
|
---|
57 | const GBM *gbm, const byte *data24
|
---|
58 | )
|
---|
59 | {
|
---|
60 | int stride24 = ((gbm->w * 3 + 3) & ~3);
|
---|
61 | int step24 = stride24 - gbm->w * 3;
|
---|
62 | FREQ *f = hist->f ;
|
---|
63 | word *ht = hist->ht;
|
---|
64 | byte rm = hist->rm;
|
---|
65 | byte gm = hist->gm;
|
---|
66 | byte bm = hist->bm;
|
---|
67 | int x, y, n_cols = hist->n_cols;
|
---|
68 |
|
---|
69 | for ( y = 0; y < gbm->h; y++, data24 += step24 )
|
---|
70 | for ( x = 0; x < gbm->w; x++ )
|
---|
71 | {
|
---|
72 | byte b = (byte) (*data24++ & bm);
|
---|
73 | byte g = (byte) (*data24++ & gm);
|
---|
74 | byte r = (byte) (*data24++ & rm);
|
---|
75 | word hc = HASH(r,g,b);
|
---|
76 | word inx;
|
---|
77 |
|
---|
78 | for ( ;; )
|
---|
79 | {
|
---|
80 | inx = ht[hc];
|
---|
81 | if ( inx == 0xffff ||
|
---|
82 | (f[inx].r == r &&
|
---|
83 | f[inx].g == g &&
|
---|
84 | f[inx].b == b) )
|
---|
85 | break;
|
---|
86 | if ( ++hc == N_HASH ) hc = 0;
|
---|
87 | }
|
---|
88 |
|
---|
89 | /* Note: loop will always be broken out of */
|
---|
90 | /* We don't allow ht to fill up above half full */
|
---|
91 |
|
---|
92 | if ( inx == 0xffff )
|
---|
93 | /* Not found in hash table */
|
---|
94 | {
|
---|
95 | if ( n_cols == N_COLS )
|
---|
96 | return FALSE;
|
---|
97 | f[n_cols].freq = (dword) 1;
|
---|
98 | f[n_cols].b = b;
|
---|
99 | f[n_cols].g = g;
|
---|
100 | f[n_cols].r = r;
|
---|
101 | ht[hc] = n_cols++;
|
---|
102 | }
|
---|
103 | else
|
---|
104 | /* Found in hash table */
|
---|
105 | /* update index inx */
|
---|
106 | f[inx].freq++;
|
---|
107 | }
|
---|
108 | hist->n_cols = n_cols;
|
---|
109 | return TRUE;
|
---|
110 | }
|
---|
111 | /*...e*/
|
---|
112 | /*...sgbm_pal_hist \45\ work out a palette from hist:0:*/
|
---|
113 | void gbm_pal_hist(
|
---|
114 | GBMHIST *hist,
|
---|
115 | GBMRGB gbmrgb[],
|
---|
116 | int n_cols_wanted
|
---|
117 | )
|
---|
118 | {
|
---|
119 | FREQ *f = hist->f;
|
---|
120 | int i;
|
---|
121 |
|
---|
122 | /* Now find the n_cols_wanted most frequently used ones */
|
---|
123 |
|
---|
124 | for ( i = 0; i < n_cols_wanted && i < hist->n_cols; i++ )
|
---|
125 | {
|
---|
126 | int j, max_j;
|
---|
127 | dword max_freq = 0;
|
---|
128 |
|
---|
129 | for ( j = 0; j < hist->n_cols; j++ )
|
---|
130 | if ( f[j].freq > max_freq )
|
---|
131 | {
|
---|
132 | max_j = j;
|
---|
133 | max_freq = f[j].freq;
|
---|
134 | }
|
---|
135 | f[max_j].nearest = (byte) i;
|
---|
136 | f[max_j].freq = (dword) 0; /* Prevent later use of f[max_j] */
|
---|
137 | gbmrgb[i].b = f[max_j].b;
|
---|
138 | gbmrgb[i].g = f[max_j].g;
|
---|
139 | gbmrgb[i].r = f[max_j].r;
|
---|
140 | }
|
---|
141 |
|
---|
142 | /* Unused palette entries will be medium grey */
|
---|
143 | for ( ; i < 0x100; i++ )
|
---|
144 | {
|
---|
145 | gbmrgb[i].r = 0x80;
|
---|
146 | gbmrgb[i].g = 0x80;
|
---|
147 | gbmrgb[i].b = 0x80;
|
---|
148 | }
|
---|
149 |
|
---|
150 | /* For the rest, find the closest one in the first n_cols_wanted */
|
---|
151 |
|
---|
152 | for ( i = 0; i < hist->n_cols; i++ )
|
---|
153 | if ( f[i].freq != (dword) 0 )
|
---|
154 | {
|
---|
155 | int j, min_j;
|
---|
156 | int min_dist = 3*256*256;
|
---|
157 |
|
---|
158 | for ( j = 0; j < n_cols_wanted; j++ )
|
---|
159 | {
|
---|
160 | int db = (int) f[i].b - (int) gbmrgb[j].b;
|
---|
161 | int dg = (int) f[i].g - (int) gbmrgb[j].g;
|
---|
162 | int dr = (int) f[i].r - (int) gbmrgb[j].r;
|
---|
163 | int dist = dr*dr + dg*dg + db*db;
|
---|
164 |
|
---|
165 | if ( dist < min_dist )
|
---|
166 | {
|
---|
167 | min_dist = dist;
|
---|
168 | min_j = j;
|
---|
169 | }
|
---|
170 | }
|
---|
171 | f[i].nearest = (byte) min_j;
|
---|
172 | }
|
---|
173 | }
|
---|
174 | /*...e*/
|
---|
175 | /*...sgbm_map_hist \45\ map bitmap data to hist palette:0:*/
|
---|
176 | void gbm_map_hist(
|
---|
177 | GBMHIST *hist,
|
---|
178 | const GBM *gbm, const byte *data24, byte *data8
|
---|
179 | )
|
---|
180 | {
|
---|
181 | int stride24 = ((gbm->w * 3 + 3) & ~3);
|
---|
182 | int step24 = stride24 - gbm->w * 3;
|
---|
183 | int stride8 = ((gbm->w + 3) & ~3);
|
---|
184 | int step8 = stride8 - gbm->w;
|
---|
185 | FREQ *f = hist->f;
|
---|
186 | word *ht = hist->ht;
|
---|
187 | byte rm = hist->rm;
|
---|
188 | byte gm = hist->gm;
|
---|
189 | byte bm = hist->bm;
|
---|
190 | int x, y;
|
---|
191 |
|
---|
192 | for ( y = 0; y < gbm->h; y++, data24 += step24, data8 += step8 )
|
---|
193 | for ( x = 0; x < gbm->w; x++ )
|
---|
194 | {
|
---|
195 | byte b = (*data24++ & bm);
|
---|
196 | byte g = (*data24++ & gm);
|
---|
197 | byte r = (*data24++ & rm);
|
---|
198 | word hc = HASH(r,g,b);
|
---|
199 | word inx;
|
---|
200 |
|
---|
201 | for ( ;; )
|
---|
202 | {
|
---|
203 | inx = ht[hc];
|
---|
204 | if ( f[inx].r == r && f[inx].g == g && f[inx].b == b )
|
---|
205 | break;
|
---|
206 | if ( ++hc == N_HASH ) hc = 0;
|
---|
207 | }
|
---|
208 |
|
---|
209 | *data8++ = f[inx].nearest;
|
---|
210 | }
|
---|
211 | }
|
---|
212 | /*...e*/
|
---|
213 | /*...sgbm_hist \45\ map single bitmap to frequency optimised palette:0:*/
|
---|
214 | /*
|
---|
215 | Determine the n_cols_wanted most frequently used colours from 24 bit data.
|
---|
216 | Can be a problem since potentially 256*256*256 possible unique colours.
|
---|
217 | Initially 8 bits green, 8 bits red, and 8 bits blue significant.
|
---|
218 | When number of colours exceeds a limit number of bits of blue reduced by 1.
|
---|
219 | Next time red, next time green, ...
|
---|
220 | Sort most n_cols_wanted most frequently used colour in order of use.
|
---|
221 | Put these in the returned palette.
|
---|
222 | Map colours from n_cols_wanted exactly to colours in palette.
|
---|
223 | For other colours, map them to the closest in the palette.
|
---|
224 | */
|
---|
225 |
|
---|
226 | BOOLEAN gbm_hist(
|
---|
227 | const GBM *gbm, const byte *data24,
|
---|
228 | GBMRGB gbmrgb[],
|
---|
229 | byte *data8,
|
---|
230 | int n_cols_wanted,
|
---|
231 | byte rm, byte gm, byte bm
|
---|
232 | )
|
---|
233 | {
|
---|
234 | GBMHIST *hist;
|
---|
235 |
|
---|
236 | for ( ;; )
|
---|
237 | {
|
---|
238 | if ( (hist = gbm_create_hist(rm, gm, bm)) == NULL )
|
---|
239 | return FALSE;
|
---|
240 |
|
---|
241 | if ( gbm_add_to_hist(hist, gbm, data24) )
|
---|
242 | break;
|
---|
243 |
|
---|
244 | gbm_delete_hist(hist);
|
---|
245 |
|
---|
246 | if ( gm > rm )
|
---|
247 | gm <<= 1;
|
---|
248 | else if ( rm > bm )
|
---|
249 | rm <<= 1;
|
---|
250 | else
|
---|
251 | bm <<= 1;
|
---|
252 | }
|
---|
253 |
|
---|
254 | /* Above loop will always be exited as if masks get rough
|
---|
255 | enough, ultimately number of unique colours < N_COLS */
|
---|
256 |
|
---|
257 | gbm_pal_hist(hist, gbmrgb, n_cols_wanted);
|
---|
258 | gbm_map_hist(hist, gbm, data24, data8);
|
---|
259 | gbm_delete_hist(hist);
|
---|
260 | return TRUE;
|
---|
261 | }
|
---|
262 | /*...e*/
|
---|