7 |
|
* G. Ward |
8 |
|
*/ |
9 |
|
|
10 |
+ |
/**************************************************************** |
11 |
+ |
1) Collect samples into a grid using the Shirley-Chiu |
12 |
+ |
angular mapping from a hemisphere to a square. |
13 |
+ |
|
14 |
+ |
2) Compute an adaptive quadtree by subdividing the grid so that |
15 |
+ |
each leaf node has at least one sample up to as many |
16 |
+ |
samples as fit nicely on a plane to within a certain |
17 |
+ |
MSE tolerance. |
18 |
+ |
|
19 |
+ |
3) Place one Gaussian lobe at each leaf node in the quadtree, |
20 |
+ |
sizing it to have a radius equal to the leaf size and |
21 |
+ |
a volume equal to the energy in that node. |
22 |
+ |
*****************************************************************/ |
23 |
+ |
|
24 |
|
#define _USE_MATH_DEFINES |
25 |
|
#include <stdio.h> |
26 |
|
#include <stdlib.h> |
29 |
|
#include "bsdfrep.h" |
30 |
|
|
31 |
|
#ifndef RSCA |
32 |
< |
#define RSCA 2.7 /* radius scaling factor (empirical) */ |
32 |
> |
#define RSCA 2.0 /* radius scaling factor (empirical) */ |
33 |
|
#endif |
34 |
< |
#ifndef MAXFRAC |
35 |
< |
#define MAXFRAC 0.5 /* maximum contribution to neighbor */ |
34 |
> |
#ifndef MAXSLOPE |
35 |
> |
#define MAXSLOPE 1000.0 /* maximum slope for smooth region */ |
36 |
|
#endif |
37 |
< |
#ifndef NNEIGH |
38 |
< |
#define NNEIGH 10 /* number of neighbors to consider */ |
37 |
> |
#ifndef SMOOTH_MSE |
38 |
> |
#define SMOOTH_MSE 5e-5 /* acceptable mean squared error */ |
39 |
|
#endif |
40 |
< |
/* our loaded grid for this incident angle */ |
40 |
> |
#ifndef SMOOTH_MSER |
41 |
> |
#define SMOOTH_MSER 0.03 /* acceptable relative MSE */ |
42 |
> |
#endif |
43 |
> |
#define MAX_RAD (GRIDRES/8) /* maximum lobe radius */ |
44 |
> |
|
45 |
> |
#define RBFALLOCB 10 /* RBF allocation block size */ |
46 |
> |
|
47 |
> |
/* loaded grid or comparison DSFs */ |
48 |
|
GRIDVAL dsf_grid[GRIDRES][GRIDRES]; |
49 |
+ |
/* allocated chrominance sums if any */ |
50 |
+ |
float (*spec_grid)[GRIDRES][GRIDRES]; |
51 |
+ |
int nspec_grid = 0; |
52 |
|
|
53 |
+ |
/* Set up visible spectrum sampling */ |
54 |
+ |
void |
55 |
+ |
set_spectral_samples(int nspec) |
56 |
+ |
{ |
57 |
+ |
if (rbf_colorimetry == RBCunknown) { |
58 |
+ |
if (nspec_grid > 0) { |
59 |
+ |
free(spec_grid); |
60 |
+ |
spec_grid = NULL; |
61 |
+ |
nspec_grid = 0; |
62 |
+ |
} |
63 |
+ |
if (nspec == 1) { |
64 |
+ |
rbf_colorimetry = RBCphotopic; |
65 |
+ |
return; |
66 |
+ |
} |
67 |
+ |
if (nspec == 3) { |
68 |
+ |
rbf_colorimetry = RBCtristimulus; |
69 |
+ |
spec_grid = (float (*)[GRIDRES][GRIDRES])calloc( |
70 |
+ |
2*GRIDRES*GRIDRES, sizeof(float) ); |
71 |
+ |
if (spec_grid == NULL) |
72 |
+ |
goto mem_error; |
73 |
+ |
nspec_grid = 2; |
74 |
+ |
return; |
75 |
+ |
} |
76 |
+ |
fprintf(stderr, |
77 |
+ |
"%s: only 1 or 3 spectral samples currently supported\n", |
78 |
+ |
progname); |
79 |
+ |
exit(1); |
80 |
+ |
} |
81 |
+ |
if (nspec != nspec_grid+1) { |
82 |
+ |
fprintf(stderr, |
83 |
+ |
"%s: number of spectral samples cannot be changed\n", |
84 |
+ |
progname); |
85 |
+ |
exit(1); |
86 |
+ |
} |
87 |
+ |
return; |
88 |
+ |
mem_error: |
89 |
+ |
fprintf(stderr, "%s: out of memory in set_spectral_samples()\n", |
90 |
+ |
progname); |
91 |
+ |
exit(1); |
92 |
+ |
} |
93 |
+ |
|
94 |
|
/* Start new DSF input grid */ |
95 |
|
void |
96 |
|
new_bsdf_data(double new_theta, double new_phi) |
98 |
|
if (!new_input_direction(new_theta, new_phi)) |
99 |
|
exit(1); |
100 |
|
memset(dsf_grid, 0, sizeof(dsf_grid)); |
101 |
+ |
if (nspec_grid > 0) |
102 |
+ |
memset(spec_grid, 0, sizeof(float)*GRIDRES*GRIDRES*nspec_grid); |
103 |
|
} |
104 |
|
|
105 |
|
/* Add BSDF data point */ |
106 |
|
void |
107 |
< |
add_bsdf_data(double theta_out, double phi_out, double val, int isDSF) |
107 |
> |
add_bsdf_data(double theta_out, double phi_out, const double val[], int isDSF) |
108 |
|
{ |
109 |
|
FVECT ovec; |
110 |
+ |
double cfact, Yval; |
111 |
|
int pos[2]; |
112 |
|
|
113 |
+ |
if (nspec_grid > 2) { |
114 |
+ |
fprintf(stderr, "%s: unsupported color space\n", progname); |
115 |
+ |
exit(1); |
116 |
+ |
} |
117 |
|
if (!output_orient) /* check output orientation */ |
118 |
|
output_orient = 1 - 2*(theta_out > 90.); |
119 |
|
else if (output_orient > 0 ^ theta_out < 90.) { |
120 |
< |
fputs("Cannot handle output angles on both sides of surface\n", |
121 |
< |
stderr); |
120 |
> |
fprintf(stderr, |
121 |
> |
"%s: cannot handle output angles on both sides of surface\n", |
122 |
> |
progname); |
123 |
|
exit(1); |
124 |
|
} |
125 |
|
ovec[2] = sin((M_PI/180.)*theta_out); |
126 |
|
ovec[0] = cos((M_PI/180.)*phi_out) * ovec[2]; |
127 |
|
ovec[1] = sin((M_PI/180.)*phi_out) * ovec[2]; |
128 |
|
ovec[2] = sqrt(1. - ovec[2]*ovec[2]); |
129 |
+ |
/* BSDF to DSF correction */ |
130 |
+ |
cfact = isDSF ? 1. : ovec[2]; |
131 |
|
|
132 |
< |
if (val <= 0) /* truncate to zero */ |
58 |
< |
val = 0; |
59 |
< |
else if (!isDSF) |
60 |
< |
val *= ovec[2]; /* convert from BSDF to DSF */ |
61 |
< |
|
132 |
> |
Yval = cfact * val[rbf_colorimetry==RBCtristimulus]; |
133 |
|
/* update BSDF histogram */ |
134 |
< |
if (val < BSDF2BIG*ovec[2] && val > BSDF2SML*ovec[2]) |
135 |
< |
++bsdf_hist[histndx(val/ovec[2])]; |
134 |
> |
if (BSDF2SML*ovec[2] < Yval && Yval < BSDF2BIG*ovec[2]) |
135 |
> |
++bsdf_hist[histndx(Yval/ovec[2])]; |
136 |
|
|
137 |
|
pos_from_vec(pos, ovec); |
138 |
|
|
139 |
< |
dsf_grid[pos[0]][pos[1]].vsum += val; |
140 |
< |
dsf_grid[pos[0]][pos[1]].nval++; |
139 |
> |
dsf_grid[pos[0]][pos[1]].sum.v += Yval; |
140 |
> |
dsf_grid[pos[0]][pos[1]].sum.n++; |
141 |
> |
/* add in X and Z values */ |
142 |
> |
if (rbf_colorimetry == RBCtristimulus) { |
143 |
> |
spec_grid[0][pos[0]][pos[1]] += cfact * val[0]; |
144 |
> |
spec_grid[1][pos[0]][pos[1]] += cfact * val[2]; |
145 |
> |
} |
146 |
|
} |
147 |
|
|
148 |
< |
/* Compute radii for non-empty bins */ |
73 |
< |
/* (distance to furthest empty bin for which non-empty bin is the closest) */ |
148 |
> |
/* Compute minimum BSDF from histogram (does not clear) */ |
149 |
|
static void |
75 |
– |
compute_radii(void) |
76 |
– |
{ |
77 |
– |
unsigned int fill_grid[GRIDRES][GRIDRES]; |
78 |
– |
unsigned short fill_cnt[GRIDRES][GRIDRES]; |
79 |
– |
FVECT ovec0, ovec1; |
80 |
– |
double ang2, lastang2; |
81 |
– |
int r, i, j, jn, ii, jj, inear, jnear; |
82 |
– |
|
83 |
– |
r = GRIDRES/2; /* proceed in zig-zag */ |
84 |
– |
for (i = 0; i < GRIDRES; i++) |
85 |
– |
for (jn = 0; jn < GRIDRES; jn++) { |
86 |
– |
j = (i&1) ? jn : GRIDRES-1-jn; |
87 |
– |
if (dsf_grid[i][j].nval) /* find empty grid pos. */ |
88 |
– |
continue; |
89 |
– |
ovec_from_pos(ovec0, i, j); |
90 |
– |
inear = jnear = -1; /* find nearest non-empty */ |
91 |
– |
lastang2 = M_PI*M_PI; |
92 |
– |
for (ii = i-r; ii <= i+r; ii++) { |
93 |
– |
if (ii < 0) continue; |
94 |
– |
if (ii >= GRIDRES) break; |
95 |
– |
for (jj = j-r; jj <= j+r; jj++) { |
96 |
– |
if (jj < 0) continue; |
97 |
– |
if (jj >= GRIDRES) break; |
98 |
– |
if (!dsf_grid[ii][jj].nval) |
99 |
– |
continue; |
100 |
– |
ovec_from_pos(ovec1, ii, jj); |
101 |
– |
ang2 = 2. - 2.*DOT(ovec0,ovec1); |
102 |
– |
if (ang2 >= lastang2) |
103 |
– |
continue; |
104 |
– |
lastang2 = ang2; |
105 |
– |
inear = ii; jnear = jj; |
106 |
– |
} |
107 |
– |
} |
108 |
– |
if (inear < 0) { |
109 |
– |
fprintf(stderr, |
110 |
– |
"%s: Could not find non-empty neighbor!\n", |
111 |
– |
progname); |
112 |
– |
exit(1); |
113 |
– |
} |
114 |
– |
ang2 = sqrt(lastang2); |
115 |
– |
r = ANG2R(ang2); /* record if > previous */ |
116 |
– |
if (r > dsf_grid[inear][jnear].crad) |
117 |
– |
dsf_grid[inear][jnear].crad = r; |
118 |
– |
/* next search radius */ |
119 |
– |
r = ang2*(2.*GRIDRES/M_PI) + 3; |
120 |
– |
} |
121 |
– |
/* blur radii over hemisphere */ |
122 |
– |
memset(fill_grid, 0, sizeof(fill_grid)); |
123 |
– |
memset(fill_cnt, 0, sizeof(fill_cnt)); |
124 |
– |
for (i = 0; i < GRIDRES; i++) |
125 |
– |
for (j = 0; j < GRIDRES; j++) { |
126 |
– |
if (!dsf_grid[i][j].crad) |
127 |
– |
continue; /* missing distance */ |
128 |
– |
r = R2ANG(dsf_grid[i][j].crad)*(2.*RSCA*GRIDRES/M_PI); |
129 |
– |
for (ii = i-r; ii <= i+r; ii++) { |
130 |
– |
if (ii < 0) continue; |
131 |
– |
if (ii >= GRIDRES) break; |
132 |
– |
for (jj = j-r; jj <= j+r; jj++) { |
133 |
– |
if (jj < 0) continue; |
134 |
– |
if (jj >= GRIDRES) break; |
135 |
– |
if ((ii-i)*(ii-i) + (jj-j)*(jj-j) > r*r) |
136 |
– |
continue; |
137 |
– |
fill_grid[ii][jj] += dsf_grid[i][j].crad; |
138 |
– |
fill_cnt[ii][jj]++; |
139 |
– |
} |
140 |
– |
} |
141 |
– |
} |
142 |
– |
/* copy back blurred radii */ |
143 |
– |
for (i = 0; i < GRIDRES; i++) |
144 |
– |
for (j = 0; j < GRIDRES; j++) |
145 |
– |
if (fill_cnt[i][j]) |
146 |
– |
dsf_grid[i][j].crad = fill_grid[i][j]/fill_cnt[i][j]; |
147 |
– |
} |
148 |
– |
|
149 |
– |
/* Cull points for more uniform distribution, leave all nval 0 or 1 */ |
150 |
– |
static void |
151 |
– |
cull_values(void) |
152 |
– |
{ |
153 |
– |
FVECT ovec0, ovec1; |
154 |
– |
double maxang, maxang2; |
155 |
– |
int i, j, ii, jj, r; |
156 |
– |
/* simple greedy algorithm */ |
157 |
– |
for (i = 0; i < GRIDRES; i++) |
158 |
– |
for (j = 0; j < GRIDRES; j++) { |
159 |
– |
if (!dsf_grid[i][j].nval) |
160 |
– |
continue; |
161 |
– |
if (!dsf_grid[i][j].crad) |
162 |
– |
continue; /* shouldn't happen */ |
163 |
– |
ovec_from_pos(ovec0, i, j); |
164 |
– |
maxang = 2.*R2ANG(dsf_grid[i][j].crad); |
165 |
– |
if (maxang > ovec0[2]) /* clamp near horizon */ |
166 |
– |
maxang = ovec0[2]; |
167 |
– |
r = maxang*(2.*GRIDRES/M_PI) + 1; |
168 |
– |
maxang2 = maxang*maxang; |
169 |
– |
for (ii = i-r; ii <= i+r; ii++) { |
170 |
– |
if (ii < 0) continue; |
171 |
– |
if (ii >= GRIDRES) break; |
172 |
– |
for (jj = j-r; jj <= j+r; jj++) { |
173 |
– |
if (jj < 0) continue; |
174 |
– |
if (jj >= GRIDRES) break; |
175 |
– |
if (!dsf_grid[ii][jj].nval) |
176 |
– |
continue; |
177 |
– |
if ((ii == i) & (jj == j)) |
178 |
– |
continue; /* don't get self-absorbed */ |
179 |
– |
ovec_from_pos(ovec1, ii, jj); |
180 |
– |
if (2. - 2.*DOT(ovec0,ovec1) >= maxang2) |
181 |
– |
continue; |
182 |
– |
/* absorb sum */ |
183 |
– |
dsf_grid[i][j].vsum += dsf_grid[ii][jj].vsum; |
184 |
– |
dsf_grid[i][j].nval += dsf_grid[ii][jj].nval; |
185 |
– |
/* keep value, though */ |
186 |
– |
dsf_grid[ii][jj].vsum /= (float)dsf_grid[ii][jj].nval; |
187 |
– |
dsf_grid[ii][jj].nval = 0; |
188 |
– |
} |
189 |
– |
} |
190 |
– |
} |
191 |
– |
/* final averaging pass */ |
192 |
– |
for (i = 0; i < GRIDRES; i++) |
193 |
– |
for (j = 0; j < GRIDRES; j++) |
194 |
– |
if (dsf_grid[i][j].nval > 1) { |
195 |
– |
dsf_grid[i][j].vsum /= (float)dsf_grid[i][j].nval; |
196 |
– |
dsf_grid[i][j].nval = 1; |
197 |
– |
} |
198 |
– |
} |
199 |
– |
|
200 |
– |
/* Compute minimum BSDF from histogram and clear it */ |
201 |
– |
static void |
150 |
|
comp_bsdf_min() |
151 |
|
{ |
152 |
< |
int cnt; |
153 |
< |
int i, target; |
152 |
> |
unsigned long cnt, target; |
153 |
> |
int i; |
154 |
|
|
155 |
|
cnt = 0; |
156 |
|
for (i = HISTLEN; i--; ) |
164 |
|
for (i = 0; cnt <= target; i++) |
165 |
|
cnt += bsdf_hist[i]; |
166 |
|
bsdf_min = histval(i-1); |
219 |
– |
memset(bsdf_hist, 0, sizeof(bsdf_hist)); |
167 |
|
} |
168 |
|
|
169 |
< |
/* Find n nearest sub-sampled neighbors to the given grid position */ |
169 |
> |
/* Determine if the given region is empty of grid samples */ |
170 |
|
static int |
171 |
< |
get_neighbors(int neigh[][2], int n, const int i, const int j) |
171 |
> |
empty_region(int x0, int x1, int y0, int y1) |
172 |
|
{ |
173 |
< |
int k = 0; |
174 |
< |
int r; |
175 |
< |
/* search concentric squares */ |
176 |
< |
for (r = 1; r < GRIDRES; r++) { |
177 |
< |
int ii, jj; |
178 |
< |
for (ii = i-r; ii <= i+r; ii++) { |
179 |
< |
int jstep = 1; |
180 |
< |
if (ii < 0) continue; |
181 |
< |
if (ii >= GRIDRES) break; |
182 |
< |
if ((i-r < ii) & (ii < i+r)) |
183 |
< |
jstep = r<<1; |
184 |
< |
for (jj = j-r; jj <= j+r; jj += jstep) { |
185 |
< |
if (jj < 0) continue; |
186 |
< |
if (jj >= GRIDRES) break; |
187 |
< |
if (dsf_grid[ii][jj].nval) { |
188 |
< |
neigh[k][0] = ii; |
189 |
< |
neigh[k][1] = jj; |
190 |
< |
if (++k >= n) |
191 |
< |
return(n); |
192 |
< |
} |
193 |
< |
} |
173 |
> |
int x, y; |
174 |
> |
|
175 |
> |
for (x = x0; x < x1; x++) |
176 |
> |
for (y = y0; y < y1; y++) |
177 |
> |
if (dsf_grid[x][y].sum.n) |
178 |
> |
return(0); |
179 |
> |
return(1); |
180 |
> |
} |
181 |
> |
|
182 |
> |
/* Determine if the given region is smooth enough to be a single lobe */ |
183 |
> |
static int |
184 |
> |
smooth_region(int x0, int x1, int y0, int y1) |
185 |
> |
{ |
186 |
> |
RREAL rMtx[3][3]; |
187 |
> |
FVECT xvec; |
188 |
> |
double A, B, C, nvs, sqerr; |
189 |
> |
int x, y, n; |
190 |
> |
/* compute planar regression */ |
191 |
> |
memset(rMtx, 0, sizeof(rMtx)); |
192 |
> |
memset(xvec, 0, sizeof(xvec)); |
193 |
> |
for (x = x0; x < x1; x++) |
194 |
> |
for (y = y0; y < y1; y++) |
195 |
> |
if ((n = dsf_grid[x][y].sum.n) > 0) { |
196 |
> |
double z = dsf_grid[x][y].sum.v; |
197 |
> |
rMtx[0][0] += x*x*(double)n; |
198 |
> |
rMtx[0][1] += x*y*(double)n; |
199 |
> |
rMtx[0][2] += x*(double)n; |
200 |
> |
rMtx[1][1] += y*y*(double)n; |
201 |
> |
rMtx[1][2] += y*(double)n; |
202 |
> |
rMtx[2][2] += (double)n; |
203 |
> |
xvec[0] += x*z; |
204 |
> |
xvec[1] += y*z; |
205 |
> |
xvec[2] += z; |
206 |
|
} |
207 |
< |
} |
208 |
< |
return(k); |
207 |
> |
rMtx[1][0] = rMtx[0][1]; |
208 |
> |
rMtx[2][0] = rMtx[0][2]; |
209 |
> |
rMtx[2][1] = rMtx[1][2]; |
210 |
> |
nvs = rMtx[2][2]; |
211 |
> |
if (SDinvXform(rMtx, rMtx) != SDEnone) |
212 |
> |
return(1); /* colinear values */ |
213 |
> |
A = DOT(rMtx[0], xvec); |
214 |
> |
B = DOT(rMtx[1], xvec); |
215 |
> |
if (A*A + B*B > MAXSLOPE*MAXSLOPE) /* too steep? */ |
216 |
> |
return(0); |
217 |
> |
C = DOT(rMtx[2], xvec); |
218 |
> |
sqerr = 0.0; /* compute mean squared error */ |
219 |
> |
for (x = x0; x < x1; x++) |
220 |
> |
for (y = y0; y < y1; y++) |
221 |
> |
if ((n = dsf_grid[x][y].sum.n) > 0) { |
222 |
> |
double d = A*x + B*y + C - dsf_grid[x][y].sum.v/n; |
223 |
> |
sqerr += n*d*d; |
224 |
> |
} |
225 |
> |
if (sqerr <= nvs*SMOOTH_MSE) /* below absolute MSE threshold? */ |
226 |
> |
return(1); |
227 |
> |
/* OR below relative MSE threshold? */ |
228 |
> |
return(sqerr*nvs <= xvec[2]*xvec[2]*SMOOTH_MSER); |
229 |
|
} |
230 |
|
|
231 |
< |
/* Adjust coded radius for the given grid position based on neighborhood */ |
231 |
> |
/* Create new lobe based on integrated samples in region */ |
232 |
|
static int |
233 |
< |
adj_coded_radius(const int i, const int j) |
233 |
> |
create_lobe(RBFVAL *rvp, int x0, int x1, int y0, int y1) |
234 |
|
{ |
235 |
< |
const double rad0 = R2ANG(dsf_grid[i][j].crad); |
236 |
< |
double currad = RSCA * rad0; |
237 |
< |
int neigh[NNEIGH][2]; |
238 |
< |
int n; |
239 |
< |
FVECT our_dir; |
235 |
> |
double vtot = 0.0; |
236 |
> |
double CIEXtot = 0.0, CIEZtot = 0.0; |
237 |
> |
int nv = 0; |
238 |
> |
double wxsum = 0.0, wysum = 0.0, wtsum = 0.0; |
239 |
> |
double rad; |
240 |
> |
int x, y; |
241 |
> |
/* compute average for region */ |
242 |
> |
for (x = x0; x < x1; x++) |
243 |
> |
for (y = y0; y < y1; y++) |
244 |
> |
if (dsf_grid[x][y].sum.n) { |
245 |
> |
const double v = dsf_grid[x][y].sum.v; |
246 |
> |
const int n = dsf_grid[x][y].sum.n; |
247 |
|
|
248 |
< |
ovec_from_pos(our_dir, i, j); |
249 |
< |
n = get_neighbors(neigh, NNEIGH, i, j); |
250 |
< |
while (n--) { |
251 |
< |
FVECT their_dir; |
252 |
< |
double max_ratio, rad_ok2; |
253 |
< |
/* check our value at neighbor */ |
254 |
< |
ovec_from_pos(their_dir, neigh[n][0], neigh[n][1]); |
255 |
< |
max_ratio = MAXFRAC * dsf_grid[neigh[n][0]][neigh[n][1]].vsum |
256 |
< |
/ dsf_grid[i][j].vsum; |
257 |
< |
if (max_ratio >= 1) |
258 |
< |
continue; |
259 |
< |
rad_ok2 = (DOT(their_dir,our_dir) - 1.)/log(max_ratio); |
260 |
< |
if (rad_ok2 >= currad*currad) |
261 |
< |
continue; /* value fraction OK */ |
262 |
< |
currad = sqrt(rad_ok2); /* else reduce lobe radius */ |
263 |
< |
if (currad <= rad0) /* limit how small we'll go */ |
264 |
< |
return(dsf_grid[i][j].crad); |
248 |
> |
if (v > 0) { |
249 |
> |
const double wt = v / (double)n; |
250 |
> |
wxsum += wt * x; |
251 |
> |
wysum += wt * y; |
252 |
> |
wtsum += wt; |
253 |
> |
} |
254 |
> |
vtot += v; |
255 |
> |
nv += n; |
256 |
> |
if (rbf_colorimetry == RBCtristimulus) { |
257 |
> |
CIEXtot += spec_grid[0][x][y]; |
258 |
> |
CIEZtot += spec_grid[1][x][y]; |
259 |
> |
} |
260 |
> |
} |
261 |
> |
if (!nv) { |
262 |
> |
fprintf(stderr, "%s: internal - missing samples in create_lobe\n", |
263 |
> |
progname); |
264 |
> |
exit(1); |
265 |
|
} |
266 |
< |
return(ANG2R(currad)); /* encode selected radius */ |
266 |
> |
if (vtot <= 0) /* only create positive lobes */ |
267 |
> |
return(0); |
268 |
> |
/* assign color */ |
269 |
> |
if (rbf_colorimetry == RBCtristimulus) { |
270 |
> |
const double df = 1.0 / (CIEXtot + vtot + CIEZtot); |
271 |
> |
C_COLOR cclr; |
272 |
> |
c_cset(&cclr, CIEXtot*df, vtot*df); |
273 |
> |
rvp->chroma = c_encodeChroma(&cclr); |
274 |
> |
} else |
275 |
> |
rvp->chroma = c_dfchroma; |
276 |
> |
/* peak value based on integral */ |
277 |
> |
vtot *= (x1-x0)*(y1-y0)*(2.*M_PI/GRIDRES/GRIDRES)/(double)nv; |
278 |
> |
rad = (RSCA/(double)GRIDRES)*(x1-x0); |
279 |
> |
rvp->peak = vtot / ((2.*M_PI) * rad*rad); |
280 |
> |
rvp->crad = ANG2R(rad); /* put peak at centroid */ |
281 |
> |
rvp->gx = (int)(wxsum/wtsum + .5); |
282 |
> |
rvp->gy = (int)(wysum/wtsum + .5); |
283 |
> |
return(1); |
284 |
|
} |
285 |
|
|
286 |
+ |
/* Recursive function to build radial basis function representation */ |
287 |
+ |
static int |
288 |
+ |
build_rbfrep(RBFVAL **arp, int *np, int x0, int x1, int y0, int y1) |
289 |
+ |
{ |
290 |
+ |
int xmid = (x0+x1)>>1; |
291 |
+ |
int ymid = (y0+y1)>>1; |
292 |
+ |
int branched[4]; |
293 |
+ |
int nadded, nleaves; |
294 |
+ |
/* need to make this a leaf? */ |
295 |
+ |
if (empty_region(x0, xmid, y0, ymid) || |
296 |
+ |
empty_region(xmid, x1, y0, ymid) || |
297 |
+ |
empty_region(x0, xmid, ymid, y1) || |
298 |
+ |
empty_region(xmid, x1, ymid, y1)) |
299 |
+ |
return(0); |
300 |
+ |
/* add children (branches+leaves) */ |
301 |
+ |
if ((branched[0] = build_rbfrep(arp, np, x0, xmid, y0, ymid)) < 0) |
302 |
+ |
return(-1); |
303 |
+ |
if ((branched[1] = build_rbfrep(arp, np, xmid, x1, y0, ymid)) < 0) |
304 |
+ |
return(-1); |
305 |
+ |
if ((branched[2] = build_rbfrep(arp, np, x0, xmid, ymid, y1)) < 0) |
306 |
+ |
return(-1); |
307 |
+ |
if ((branched[3] = build_rbfrep(arp, np, xmid, x1, ymid, y1)) < 0) |
308 |
+ |
return(-1); |
309 |
+ |
nadded = branched[0] + branched[1] + branched[2] + branched[3]; |
310 |
+ |
nleaves = !branched[0] + !branched[1] + !branched[2] + !branched[3]; |
311 |
+ |
if (!nleaves) /* nothing but branches? */ |
312 |
+ |
return(nadded); |
313 |
+ |
/* combine 4 leaves into 1? */ |
314 |
+ |
if ((nleaves == 4) & (x1-x0 <= MAX_RAD) && |
315 |
+ |
smooth_region(x0, x1, y0, y1)) |
316 |
+ |
return(0); |
317 |
+ |
/* need more array space? */ |
318 |
+ |
if ((*np+nleaves-1)>>RBFALLOCB != (*np-1)>>RBFALLOCB) { |
319 |
+ |
*arp = (RBFVAL *)realloc(*arp, |
320 |
+ |
sizeof(RBFVAL)*(*np+nleaves-1+(1<<RBFALLOCB))); |
321 |
+ |
if (*arp == NULL) |
322 |
+ |
return(-1); |
323 |
+ |
} |
324 |
+ |
/* create lobes for leaves */ |
325 |
+ |
if (!branched[0] && create_lobe(*arp+*np, x0, xmid, y0, ymid)) { |
326 |
+ |
++(*np); ++nadded; |
327 |
+ |
} |
328 |
+ |
if (!branched[1] && create_lobe(*arp+*np, xmid, x1, y0, ymid)) { |
329 |
+ |
++(*np); ++nadded; |
330 |
+ |
} |
331 |
+ |
if (!branched[2] && create_lobe(*arp+*np, x0, xmid, ymid, y1)) { |
332 |
+ |
++(*np); ++nadded; |
333 |
+ |
} |
334 |
+ |
if (!branched[3] && create_lobe(*arp+*np, xmid, x1, ymid, y1)) { |
335 |
+ |
++(*np); ++nadded; |
336 |
+ |
} |
337 |
+ |
return(nadded); |
338 |
+ |
} |
339 |
+ |
|
340 |
|
/* Count up filled nodes and build RBF representation from current grid */ |
341 |
|
RBFNODE * |
342 |
< |
make_rbfrep(void) |
342 |
> |
make_rbfrep() |
343 |
|
{ |
287 |
– |
int niter = 16; |
288 |
– |
double lastVar, thisVar = 100.; |
289 |
– |
int nn; |
344 |
|
RBFNODE *newnode; |
345 |
< |
RBFVAL *itera; |
346 |
< |
int i, j; |
293 |
< |
/* compute RBF radii */ |
294 |
< |
compute_radii(); |
295 |
< |
/* coagulate lobes */ |
296 |
< |
cull_values(); |
297 |
< |
nn = 0; /* count selected bins */ |
298 |
< |
for (i = 0; i < GRIDRES; i++) |
299 |
< |
for (j = 0; j < GRIDRES; j++) |
300 |
< |
nn += dsf_grid[i][j].nval; |
345 |
> |
RBFVAL *rbfarr; |
346 |
> |
int nn; |
347 |
|
/* compute minimum BSDF */ |
348 |
|
comp_bsdf_min(); |
349 |
< |
/* allocate RBF array */ |
350 |
< |
newnode = (RBFNODE *)malloc(sizeof(RBFNODE) + sizeof(RBFVAL)*(nn-1)); |
349 |
> |
/* create RBF node list */ |
350 |
> |
rbfarr = NULL; nn = 0; |
351 |
> |
if (build_rbfrep(&rbfarr, &nn, 0, GRIDRES, 0, GRIDRES) <= 0) { |
352 |
> |
if (nn) |
353 |
> |
goto memerr; |
354 |
> |
fprintf(stderr, |
355 |
> |
"%s: warning - skipping bad incidence (%.1f,%.1f)\n", |
356 |
> |
progname, theta_in_deg, phi_in_deg); |
357 |
> |
return(NULL); |
358 |
> |
} |
359 |
> |
/* (re)allocate RBF array */ |
360 |
> |
newnode = (RBFNODE *)realloc(rbfarr, |
361 |
> |
sizeof(RBFNODE) + sizeof(RBFVAL)*(nn-1)); |
362 |
|
if (newnode == NULL) |
363 |
|
goto memerr; |
364 |
+ |
/* copy computed lobes into RBF node */ |
365 |
+ |
memmove(newnode->rbfa, newnode, sizeof(RBFVAL)*nn); |
366 |
|
newnode->ord = -1; |
367 |
|
newnode->next = NULL; |
368 |
|
newnode->ejl = NULL; |
370 |
|
newnode->invec[0] = cos((M_PI/180.)*phi_in_deg)*newnode->invec[2]; |
371 |
|
newnode->invec[1] = sin((M_PI/180.)*phi_in_deg)*newnode->invec[2]; |
372 |
|
newnode->invec[2] = input_orient*sqrt(1. - newnode->invec[2]*newnode->invec[2]); |
373 |
< |
newnode->vtotal = 0; |
373 |
> |
newnode->vtotal = .0; |
374 |
|
newnode->nrbf = nn; |
375 |
< |
nn = 0; /* fill RBF array */ |
376 |
< |
for (i = 0; i < GRIDRES; i++) |
377 |
< |
for (j = 0; j < GRIDRES; j++) |
319 |
< |
if (dsf_grid[i][j].nval) { |
320 |
< |
newnode->rbfa[nn].peak = dsf_grid[i][j].vsum; |
321 |
< |
newnode->rbfa[nn].crad = adj_coded_radius(i, j); |
322 |
< |
newnode->rbfa[nn].gx = i; |
323 |
< |
newnode->rbfa[nn].gy = j; |
324 |
< |
++nn; |
325 |
< |
} |
326 |
< |
/* iterate to improve interpolation accuracy */ |
327 |
< |
itera = (RBFVAL *)malloc(sizeof(RBFVAL)*newnode->nrbf); |
328 |
< |
if (itera == NULL) |
329 |
< |
goto memerr; |
330 |
< |
memcpy(itera, newnode->rbfa, sizeof(RBFVAL)*newnode->nrbf); |
331 |
< |
do { |
332 |
< |
double dsum = 0, dsum2 = 0; |
333 |
< |
nn = 0; |
334 |
< |
for (i = 0; i < GRIDRES; i++) |
335 |
< |
for (j = 0; j < GRIDRES; j++) |
336 |
< |
if (dsf_grid[i][j].nval) { |
337 |
< |
FVECT odir; |
338 |
< |
double corr; |
339 |
< |
ovec_from_pos(odir, i, j); |
340 |
< |
itera[nn++].peak *= corr = |
341 |
< |
dsf_grid[i][j].vsum / |
342 |
< |
eval_rbfrep(newnode, odir); |
343 |
< |
dsum += 1. - corr; |
344 |
< |
dsum2 += (1.-corr)*(1.-corr); |
345 |
< |
} |
346 |
< |
memcpy(newnode->rbfa, itera, sizeof(RBFVAL)*newnode->nrbf); |
347 |
< |
lastVar = thisVar; |
348 |
< |
thisVar = dsum2/(double)nn; |
375 |
> |
/* compute sum for normalization */ |
376 |
> |
while (nn-- > 0) |
377 |
> |
newnode->vtotal += rbf_volume(&newnode->rbfa[nn]); |
378 |
|
#ifdef DEBUG |
379 |
< |
fprintf(stderr, "Avg., RMS error: %.1f%% %.1f%%\n", |
351 |
< |
100.*dsum/(double)nn, |
352 |
< |
100.*sqrt(thisVar)); |
353 |
< |
#endif |
354 |
< |
} while (--niter > 0 && lastVar-thisVar > 0.02*lastVar); |
355 |
< |
|
356 |
< |
free(itera); |
357 |
< |
nn = 0; /* compute sum for normalization */ |
358 |
< |
while (nn < newnode->nrbf) |
359 |
< |
newnode->vtotal += rbf_volume(&newnode->rbfa[nn++]); |
360 |
< |
#ifdef DEBUG |
379 |
> |
fprintf(stderr, "Built RBF with %d lobes\n", newnode->nrbf); |
380 |
|
fprintf(stderr, "Integrated DSF at (%.1f,%.1f) deg. is %.2f\n", |
381 |
|
get_theta180(newnode->invec), get_phi360(newnode->invec), |
382 |
|
newnode->vtotal); |
383 |
|
#endif |
384 |
|
insert_dsf(newnode); |
366 |
– |
|
385 |
|
return(newnode); |
386 |
|
memerr: |
387 |
|
fprintf(stderr, "%s: Out of memory in make_rbfrep()\n", progname); |