1 |
#ifndef lint |
2 |
static const char RCSid[] = "$Id$"; |
3 |
#endif |
4 |
/* |
5 |
* Allocate and control dynamic color table. |
6 |
* |
7 |
* We start off with a uniform partition of color space. |
8 |
* As pixels are sent to the frame buffer, a histogram is built. |
9 |
* When a new color table is requested, the histogram is used |
10 |
* to make a pseudo-optimal partition, after which the |
11 |
* histogram is cleared. This algorithm |
12 |
* performs only as well as the next drawing's color |
13 |
* distribution is correlated to the last. |
14 |
* |
15 |
* This module is essentially identical to src/rt/colortab.c, |
16 |
* except there is no color mapping, since the tm library is used. |
17 |
*/ |
18 |
|
19 |
#include "standard.h" |
20 |
#include "color.h" |
21 |
/* histogram resolution */ |
22 |
#define NRED 24 |
23 |
#define NGRN 32 |
24 |
#define NBLU 16 |
25 |
#define HMAX NGRN |
26 |
/* minimum box count for adaptive partition */ |
27 |
#define MINSAMP 7 |
28 |
/* maximum distance^2 before color reassign */ |
29 |
#define MAXDST2 12 |
30 |
/* color partition tree */ |
31 |
#define CNODE short |
32 |
#define set_branch(p,c) ((c)<<2|(p)) |
33 |
#define set_pval(pv) ((pv)<<2|3) |
34 |
#define is_branch(cn) (((cn)&3)!=3) |
35 |
#define is_pval(cn) (((cn)&3)==3) |
36 |
#define part(cn) ((cn)>>2) |
37 |
#define prim(cn) ((cn)&3) |
38 |
#define pval(cn) ((cn)>>2) |
39 |
/* our color table */ |
40 |
static struct tabent { |
41 |
long sum[3]; /* sum of colors using this entry */ |
42 |
int n; /* number of colors */ |
43 |
BYTE ent[3]; /* current table value */ |
44 |
} *clrtab = NULL; |
45 |
/* color cube partition */ |
46 |
static CNODE *ctree = NULL; |
47 |
/* histogram of colors used */ |
48 |
static unsigned short histo[NRED][NGRN][NBLU]; |
49 |
/* initial color cube boundary */ |
50 |
static int CLRCUBE[3][2] = {{0,NRED},{0,NGRN},{0,NBLU}}; |
51 |
|
52 |
static int split(), cut(); |
53 |
|
54 |
|
55 |
int |
56 |
new_ctab(ncolors) /* start new color table with max ncolors */ |
57 |
int ncolors; |
58 |
{ |
59 |
int treesize; |
60 |
|
61 |
if (ncolors < 1) |
62 |
return(0); |
63 |
/* free old tables */ |
64 |
if (clrtab != NULL) |
65 |
free((void *)clrtab); |
66 |
if (ctree != NULL) |
67 |
free((void *)ctree); |
68 |
/* get new tables */ |
69 |
for (treesize = 1; treesize < ncolors; treesize <<= 1) |
70 |
; |
71 |
treesize <<= 1; |
72 |
clrtab = (struct tabent *)calloc(ncolors, sizeof(struct tabent)); |
73 |
ctree = (CNODE *)malloc(treesize*sizeof(CNODE)); |
74 |
if (clrtab == NULL || ctree == NULL) |
75 |
return(0); |
76 |
/* partition color space */ |
77 |
cut(ctree, 0, CLRCUBE, 0, ncolors); |
78 |
/* clear histogram */ |
79 |
bzero((char *)histo, sizeof(histo)); |
80 |
/* return number of colors used */ |
81 |
return(ncolors); |
82 |
} |
83 |
|
84 |
|
85 |
int |
86 |
get_pixel(rgb, set_pixel) /* get pixel for color */ |
87 |
BYTE rgb[3]; |
88 |
int (*set_pixel)(); |
89 |
{ |
90 |
extern char errmsg[]; |
91 |
int r, g, b; |
92 |
int cv[3]; |
93 |
register CNODE *tp; |
94 |
register int h; |
95 |
/* get desired color */ |
96 |
r = rgb[RED]; |
97 |
g = rgb[GRN]; |
98 |
b = rgb[BLU]; |
99 |
/* reduce resolution */ |
100 |
cv[RED] = (r*NRED)>>8; |
101 |
cv[GRN] = (g*NGRN)>>8; |
102 |
cv[BLU] = (b*NBLU)>>8; |
103 |
/* add to histogram */ |
104 |
histo[cv[RED]][cv[GRN]][cv[BLU]]++; |
105 |
/* find pixel in tree */ |
106 |
for (tp = ctree, h = 0; is_branch(*tp); h++) |
107 |
if (cv[prim(*tp)] < part(*tp)) |
108 |
tp += 1<<h; /* left branch */ |
109 |
else |
110 |
tp += 1<<(h+1); /* right branch */ |
111 |
h = pval(*tp); |
112 |
/* add to color table */ |
113 |
clrtab[h].sum[RED] += r; |
114 |
clrtab[h].sum[GRN] += g; |
115 |
clrtab[h].sum[BLU] += b; |
116 |
clrtab[h].n++; |
117 |
/* recompute average */ |
118 |
r = clrtab[h].sum[RED] / clrtab[h].n; |
119 |
g = clrtab[h].sum[GRN] / clrtab[h].n; |
120 |
b = clrtab[h].sum[BLU] / clrtab[h].n; |
121 |
/* check for movement */ |
122 |
if (clrtab[h].n == 1 || |
123 |
(r-clrtab[h].ent[RED])*(r-clrtab[h].ent[RED]) + |
124 |
(g-clrtab[h].ent[GRN])*(g-clrtab[h].ent[GRN]) + |
125 |
(b-clrtab[h].ent[BLU])*(b-clrtab[h].ent[BLU]) > MAXDST2) { |
126 |
clrtab[h].ent[RED] = r; |
127 |
clrtab[h].ent[GRN] = g; /* reassign pixel */ |
128 |
clrtab[h].ent[BLU] = b; |
129 |
#ifdef DEBUG |
130 |
sprintf(errmsg, "pixel %d = (%d,%d,%d) (%d refs)\n", |
131 |
h, r, g, b, clrtab[h].n); |
132 |
eputs(errmsg); |
133 |
#endif |
134 |
(*set_pixel)(h, r, g, b); |
135 |
} |
136 |
return(h); /* return pixel value */ |
137 |
} |
138 |
|
139 |
|
140 |
static |
141 |
cut(tree, level, box, c0, c1) /* partition color space */ |
142 |
register CNODE *tree; |
143 |
int level; |
144 |
register int box[3][2]; |
145 |
int c0, c1; |
146 |
{ |
147 |
int kb[3][2]; |
148 |
|
149 |
if (c1-c0 <= 1) { /* assign pixel */ |
150 |
*tree = set_pval(c0); |
151 |
return; |
152 |
} |
153 |
/* split box */ |
154 |
*tree = split(box); |
155 |
bcopy((char *)box, (char *)kb, sizeof(kb)); |
156 |
/* do left (lesser) branch */ |
157 |
kb[prim(*tree)][1] = part(*tree); |
158 |
cut(tree+(1<<level), level+1, kb, c0, (c0+c1)>>1); |
159 |
/* do right branch */ |
160 |
kb[prim(*tree)][0] = part(*tree); |
161 |
kb[prim(*tree)][1] = box[prim(*tree)][1]; |
162 |
cut(tree+(1<<(level+1)), level+1, kb, (c0+c1)>>1, c1); |
163 |
} |
164 |
|
165 |
|
166 |
static int |
167 |
split(box) /* find median cut for box */ |
168 |
register int box[3][2]; |
169 |
{ |
170 |
#define c0 r |
171 |
register int r, g, b; |
172 |
int pri; |
173 |
long t[HMAX], med; |
174 |
/* find dominant axis */ |
175 |
pri = RED; |
176 |
if (box[GRN][1]-box[GRN][0] > box[pri][1]-box[pri][0]) |
177 |
pri = GRN; |
178 |
if (box[BLU][1]-box[BLU][0] > box[pri][1]-box[pri][0]) |
179 |
pri = BLU; |
180 |
/* sum histogram over box */ |
181 |
med = 0; |
182 |
switch (pri) { |
183 |
case RED: |
184 |
for (r = box[RED][0]; r < box[RED][1]; r++) { |
185 |
t[r] = 0; |
186 |
for (g = box[GRN][0]; g < box[GRN][1]; g++) |
187 |
for (b = box[BLU][0]; b < box[BLU][1]; b++) |
188 |
t[r] += histo[r][g][b]; |
189 |
med += t[r]; |
190 |
} |
191 |
break; |
192 |
case GRN: |
193 |
for (g = box[GRN][0]; g < box[GRN][1]; g++) { |
194 |
t[g] = 0; |
195 |
for (b = box[BLU][0]; b < box[BLU][1]; b++) |
196 |
for (r = box[RED][0]; r < box[RED][1]; r++) |
197 |
t[g] += histo[r][g][b]; |
198 |
med += t[g]; |
199 |
} |
200 |
break; |
201 |
case BLU: |
202 |
for (b = box[BLU][0]; b < box[BLU][1]; b++) { |
203 |
t[b] = 0; |
204 |
for (r = box[RED][0]; r < box[RED][1]; r++) |
205 |
for (g = box[GRN][0]; g < box[GRN][1]; g++) |
206 |
t[b] += histo[r][g][b]; |
207 |
med += t[b]; |
208 |
} |
209 |
break; |
210 |
} |
211 |
if (med < MINSAMP) /* if too sparse, split at midpoint */ |
212 |
return(set_branch(pri,(box[pri][0]+box[pri][1])>>1)); |
213 |
/* find median position */ |
214 |
med >>= 1; |
215 |
for (c0 = box[pri][0]; med > 0; c0++) |
216 |
med -= t[c0]; |
217 |
if (c0 > (box[pri][0]+box[pri][1])>>1) /* if past the midpoint */ |
218 |
c0--; /* part left of median */ |
219 |
return(set_branch(pri,c0)); |
220 |
#undef c0 |
221 |
} |