Parent Directory | Revision Log
Revision 1.6 - (view) (download)
1 : | edgomez | 1.2 | /***************************************************************************** |
2 : | * | ||
3 : | * XVID MPEG-4 VIDEO CODEC | ||
4 : | * - Global Motion Estimation - | ||
5 : | * | ||
6 : | * Copyright(C) 2003 Christoph Lampert <gruel@web.de> | ||
7 : | * | ||
8 : | * This program is free software ; you can redistribute it and/or modify | ||
9 : | * it under the terms of the GNU General Public License as published by | ||
10 : | * the Free Software Foundation ; either version 2 of the License, or | ||
11 : | * (at your option) any later version. | ||
12 : | * | ||
13 : | * This program is distributed in the hope that it will be useful, | ||
14 : | * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
15 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
16 : | * GNU General Public License for more details. | ||
17 : | * | ||
18 : | * You should have received a copy of the GNU General Public License | ||
19 : | * along with this program ; if not, write to the Free Software | ||
20 : | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
21 : | * | ||
22 : | Isibaar | 1.6 | * $Id: estimation_gmc.c,v 1.5 2004/12/05 04:53:01 syskin Exp $ |
23 : | edgomez | 1.2 | * |
24 : | ****************************************************************************/ | ||
25 : | |||
26 : | #include <assert.h> | ||
27 : | #include <stdio.h> | ||
28 : | #include <stdlib.h> | ||
29 : | #include <string.h> | ||
30 : | #include <math.h> | ||
31 : | |||
32 : | #include "../encoder.h" | ||
33 : | #include "../prediction/mbprediction.h" | ||
34 : | #include "estimation.h" | ||
35 : | #include "motion.h" | ||
36 : | #include "sad.h" | ||
37 : | #include "gmc.h" | ||
38 : | #include "../utils/emms.h" | ||
39 : | #include "motion_inlines.h" | ||
40 : | |||
41 : | static void | ||
42 : | CheckCandidate16I(const int x, const int y, SearchData * const data, const unsigned int Direction) | ||
43 : | { | ||
44 : | int sad; | ||
45 : | const uint8_t * Reference; | ||
46 : | |||
47 : | if ( (x > data->max_dx) || ( x < data->min_dx) | ||
48 : | || (y > data->max_dy) || (y < data->min_dy) ) return; | ||
49 : | |||
50 : | Reference = GetReference(x, y, data); | ||
51 : | |||
52 : | sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); | ||
53 : | |||
54 : | if (sad < data->iMinSAD[0]) { | ||
55 : | data->iMinSAD[0] = sad; | ||
56 : | data->currentMV[0].x = x; data->currentMV[0].y = y; | ||
57 : | data->dir = Direction; | ||
58 : | } | ||
59 : | } | ||
60 : | |||
61 : | static __inline void | ||
62 : | GMEanalyzeMB ( const uint8_t * const pCur, | ||
63 : | const uint8_t * const pRef, | ||
64 : | const uint8_t * const pRefH, | ||
65 : | const uint8_t * const pRefV, | ||
66 : | const uint8_t * const pRefHV, | ||
67 : | const int x, | ||
68 : | const int y, | ||
69 : | const MBParam * const pParam, | ||
70 : | MACROBLOCK * const pMBs, | ||
71 : | Isibaar | 1.6 | SearchData * const Data, |
72 : | const int bound) | ||
73 : | edgomez | 1.2 | { |
74 : | |||
75 : | MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width]; | ||
76 : | |||
77 : | Data->iMinSAD[0] = MV_MAX_ERROR; | ||
78 : | |||
79 : | Isibaar | 1.6 | Data->predMV = get_pmv2(pMBs, pParam->mb_width, bound, x, y, 0); |
80 : | edgomez | 1.2 | |
81 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, | ||
82 : | syskin | 1.5 | pParam->width, pParam->height, 16, 1); |
83 : | edgomez | 1.2 | |
84 : | Data->Cur = pCur + 16*(x + y * pParam->edged_width); | ||
85 : | Data->RefP[0] = pRef + 16*(x + y * pParam->edged_width); | ||
86 : | Data->RefP[1] = pRefV + 16*(x + y * pParam->edged_width); | ||
87 : | Data->RefP[2] = pRefH + 16*(x + y * pParam->edged_width); | ||
88 : | Data->RefP[3] = pRefHV + 16*(x + y * pParam->edged_width); | ||
89 : | |||
90 : | Data->currentMV[0].x = Data->currentMV[0].y = 0; | ||
91 : | CheckCandidate16I(0, 0, Data, 255); | ||
92 : | |||
93 : | if ( (Data->predMV.x !=0) || (Data->predMV.y != 0) ) | ||
94 : | CheckCandidate16I(Data->predMV.x, Data->predMV.y, Data, 255); | ||
95 : | |||
96 : | xvid_me_DiamondSearch(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255, CheckCandidate16I); | ||
97 : | |||
98 : | syskin | 1.4 | xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidate16I, 0); |
99 : | edgomez | 1.2 | |
100 : | |||
101 : | /* for QPel halfpel positions are worse than in halfpel mode :( */ | ||
102 : | /* if (Data->qpel) { | ||
103 : | Data->currentQMV->x = 2*Data->currentMV->x; | ||
104 : | Data->currentQMV->y = 2*Data->currentMV->y; | ||
105 : | Data->qpel_precision = 1; | ||
106 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, | ||
107 : | pParam->width, pParam->height, iFcode, 2, 0); | ||
108 : | SubpelRefine(Data); | ||
109 : | } | ||
110 : | */ | ||
111 : | |||
112 : | pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; | ||
113 : | pMB->sad16 = Data->iMinSAD[0]; | ||
114 : | pMB->mode = MODE_INTER; | ||
115 : | syskin | 1.5 | pMB->sad16 += 10*d_mv_bits(pMB->mvs[0].x, pMB->mvs[0].y, Data->predMV, Data->iFcode, 0); |
116 : | edgomez | 1.2 | return; |
117 : | } | ||
118 : | |||
119 : | |||
120 : | void | ||
121 : | GMEanalysis(const MBParam * const pParam, | ||
122 : | const FRAMEINFO * const current, | ||
123 : | const FRAMEINFO * const reference, | ||
124 : | const IMAGE * const pRefH, | ||
125 : | const IMAGE * const pRefV, | ||
126 : | Isibaar | 1.6 | const IMAGE * const pRefHV, |
127 : | const int num_slices) | ||
128 : | edgomez | 1.2 | { |
129 : | uint32_t x, y; | ||
130 : | MACROBLOCK * const pMBs = current->mbs; | ||
131 : | const IMAGE * const pCurrent = ¤t->image; | ||
132 : | const IMAGE * const pReference = &reference->image; | ||
133 : | Isibaar | 1.6 | int bound = 0; |
134 : | const uint32_t mb_width = pParam->mb_width; | ||
135 : | const uint32_t mb_height = pParam->mb_height; | ||
136 : | edgomez | 1.2 | |
137 : | SearchData Data; | ||
138 : | memset(&Data, 0, sizeof(SearchData)); | ||
139 : | |||
140 : | Data.iEdgedWidth = pParam->edged_width; | ||
141 : | Data.rounding = pParam->m_rounding_type; | ||
142 : | |||
143 : | Data.iFcode = current->fcode; | ||
144 : | |||
145 : | if (sadInit) (*sadInit) (); | ||
146 : | |||
147 : | for (y = 0; y < pParam->mb_height; y ++) { | ||
148 : | Isibaar | 1.6 | bound = mb_width * ((((y*num_slices) / mb_height) * mb_height + (num_slices-1))/ num_slices); |
149 : | edgomez | 1.2 | for (x = 0; x < pParam->mb_width; x ++) { |
150 : | Isibaar | 1.6 | GMEanalyzeMB(pCurrent->y, pReference->y, pRefH->y, pRefV->y, pRefHV->y, x, y, pParam, pMBs, &Data, bound); |
151 : | edgomez | 1.2 | } |
152 : | } | ||
153 : | return; | ||
154 : | } | ||
155 : | |||
156 : | WARPPOINTS | ||
157 : | GlobalMotionEst(MACROBLOCK * const pMBs, | ||
158 : | const MBParam * const pParam, | ||
159 : | const FRAMEINFO * const current, | ||
160 : | const FRAMEINFO * const reference, | ||
161 : | const IMAGE * const pRefH, | ||
162 : | const IMAGE * const pRefV, | ||
163 : | Isibaar | 1.6 | const IMAGE * const pRefHV, |
164 : | const int num_slices) | ||
165 : | edgomez | 1.2 | { |
166 : | |||
167 : | const int deltax=8; /* upper bound for difference between a MV and it's neighbour MVs */ | ||
168 : | const int deltay=8; | ||
169 : | const unsigned int gradx=512; /* lower bound for gradient in MB (ignore "flat" blocks) */ | ||
170 : | const unsigned int grady=512; | ||
171 : | |||
172 : | double sol[4] = { 0., 0., 0., 0. }; | ||
173 : | |||
174 : | WARPPOINTS gmc; | ||
175 : | |||
176 : | uint32_t mx, my; | ||
177 : | |||
178 : | int MBh = pParam->mb_height; | ||
179 : | int MBw = pParam->mb_width; | ||
180 : | const int minblocks = 9; /* was = /MBh*MBw/32+3 */ /* just some reasonable number 3% + 3 */ | ||
181 : | const int maxblocks = MBh*MBw/4; /* just some reasonable number 3% + 3 */ | ||
182 : | |||
183 : | int num=0; | ||
184 : | int oldnum; | ||
185 : | |||
186 : | gmc.duv[0].x = gmc.duv[0].y = gmc.duv[1].x = gmc.duv[1].y = gmc.duv[2].x = gmc.duv[2].y = 0; | ||
187 : | |||
188 : | Isibaar | 1.6 | GMEanalysis(pParam,current, reference, pRefH, pRefV, pRefHV, num_slices); |
189 : | edgomez | 1.2 | |
190 : | /* block based ME isn't done, yet, so do a quick presearch */ | ||
191 : | |||
192 : | /* filter mask of all blocks */ | ||
193 : | |||
194 : | for (my = 0; my < (uint32_t)MBh; my++) | ||
195 : | for (mx = 0; mx < (uint32_t)MBw; mx++) | ||
196 : | { | ||
197 : | const int mbnum = mx + my * MBw; | ||
198 : | pMBs[mbnum].mcsel = 0; | ||
199 : | } | ||
200 : | |||
201 : | |||
202 : | for (my = 1; my < (uint32_t)MBh-1; my++) /* ignore boundary blocks */ | ||
203 : | for (mx = 1; mx < (uint32_t)MBw-1; mx++) /* theirs MVs are often wrong */ | ||
204 : | { | ||
205 : | const int mbnum = mx + my * MBw; | ||
206 : | MACROBLOCK *const pMB = &pMBs[mbnum]; | ||
207 : | const VECTOR mv = pMB->mvs[0]; | ||
208 : | |||
209 : | /* don't use object boundaries */ | ||
210 : | if ( (abs(mv.x - (pMB-1)->mvs[0].x) < deltax) | ||
211 : | && (abs(mv.y - (pMB-1)->mvs[0].y) < deltay) | ||
212 : | && (abs(mv.x - (pMB+1)->mvs[0].x) < deltax) | ||
213 : | && (abs(mv.y - (pMB+1)->mvs[0].y) < deltay) | ||
214 : | && (abs(mv.x - (pMB-MBw)->mvs[0].x) < deltax) | ||
215 : | && (abs(mv.y - (pMB-MBw)->mvs[0].y) < deltay) | ||
216 : | && (abs(mv.x - (pMB+MBw)->mvs[0].x) < deltax) | ||
217 : | && (abs(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) | ||
218 : | { const int iEdgedWidth = pParam->edged_width; | ||
219 : | const uint8_t *const pCur = current->image.y + 16*(my*iEdgedWidth + mx); | ||
220 : | if ( (sad16 ( pCur, pCur+1 , iEdgedWidth, 65536) >= gradx ) | ||
221 : | && (sad16 ( pCur, pCur+iEdgedWidth, iEdgedWidth, 65536) >= grady ) ) | ||
222 : | { pMB->mcsel = 1; | ||
223 : | num++; | ||
224 : | } | ||
225 : | |||
226 : | /* only use "structured" blocks */ | ||
227 : | } | ||
228 : | } | ||
229 : | emms(); | ||
230 : | |||
231 : | /* further filtering would be possible, but during iteration, remaining | ||
232 : | outliers usually are removed, too */ | ||
233 : | |||
234 : | if (num>= minblocks) | ||
235 : | do { /* until convergence */ | ||
236 : | double DtimesF[4]; | ||
237 : | double a,b,c,n,invdenom; | ||
238 : | double meanx,meany; | ||
239 : | |||
240 : | a = b = c = n = 0; | ||
241 : | DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.; | ||
242 : | for (my = 1; my < (uint32_t)MBh-1; my++) | ||
243 : | for (mx = 1; mx < (uint32_t)MBw-1; mx++) | ||
244 : | { | ||
245 : | const int mbnum = mx + my * MBw; | ||
246 : | const VECTOR mv = pMBs[mbnum].mvs[0]; | ||
247 : | |||
248 : | if (!pMBs[mbnum].mcsel) | ||
249 : | continue; | ||
250 : | |||
251 : | n++; | ||
252 : | a += 16*mx+8; | ||
253 : | b += 16*my+8; | ||
254 : | c += (16*mx+8)*(16*mx+8)+(16*my+8)*(16*my+8); | ||
255 : | |||
256 : | DtimesF[0] += (double)mv.x; | ||
257 : | DtimesF[1] += (double)mv.x*(16*mx+8) + (double)mv.y*(16*my+8); | ||
258 : | DtimesF[2] += (double)mv.x*(16*my+8) - (double)mv.y*(16*mx+8); | ||
259 : | DtimesF[3] += (double)mv.y; | ||
260 : | } | ||
261 : | |||
262 : | invdenom = a*a+b*b-c*n; | ||
263 : | |||
264 : | /* Solve the system: sol = (D'*E*D)^{-1} D'*E*F */ | ||
265 : | /* D'*E*F has been calculated in the same loop as matrix */ | ||
266 : | |||
267 : | sol[0] = -c*DtimesF[0] + a*DtimesF[1] + b*DtimesF[2]; | ||
268 : | sol[1] = a*DtimesF[0] - n*DtimesF[1] + b*DtimesF[3]; | ||
269 : | sol[2] = b*DtimesF[0] - n*DtimesF[2] - a*DtimesF[3]; | ||
270 : | sol[3] = b*DtimesF[1] - a*DtimesF[2] - c*DtimesF[3]; | ||
271 : | |||
272 : | sol[0] /= invdenom; | ||
273 : | sol[1] /= invdenom; | ||
274 : | sol[2] /= invdenom; | ||
275 : | sol[3] /= invdenom; | ||
276 : | |||
277 : | meanx = meany = 0.; | ||
278 : | oldnum = 0; | ||
279 : | for (my = 1; my < (uint32_t)MBh-1; my++) | ||
280 : | for (mx = 1; mx < (uint32_t)MBw-1; mx++) | ||
281 : | { | ||
282 : | const int mbnum = mx + my * MBw; | ||
283 : | const VECTOR mv = pMBs[mbnum].mvs[0]; | ||
284 : | |||
285 : | if (!pMBs[mbnum].mcsel) | ||
286 : | continue; | ||
287 : | |||
288 : | oldnum++; | ||
289 : | meanx += fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - (double)mv.x ); | ||
290 : | meany += fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - (double)mv.y ); | ||
291 : | } | ||
292 : | |||
293 : | if (4*meanx > oldnum) /* better fit than 0.25 (=1/4pel) is useless */ | ||
294 : | meanx /= oldnum; | ||
295 : | else | ||
296 : | meanx = 0.25; | ||
297 : | |||
298 : | if (4*meany > oldnum) | ||
299 : | meany /= oldnum; | ||
300 : | else | ||
301 : | meany = 0.25; | ||
302 : | |||
303 : | num = 0; | ||
304 : | for (my = 0; my < (uint32_t)MBh; my++) | ||
305 : | for (mx = 0; mx < (uint32_t)MBw; mx++) | ||
306 : | { | ||
307 : | const int mbnum = mx + my * MBw; | ||
308 : | const VECTOR mv = pMBs[mbnum].mvs[0]; | ||
309 : | |||
310 : | if (!pMBs[mbnum].mcsel) | ||
311 : | continue; | ||
312 : | |||
313 : | if ( ( fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - (double)mv.x ) > meanx ) | ||
314 : | || ( fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - (double)mv.y ) > meany ) ) | ||
315 : | pMBs[mbnum].mcsel=0; | ||
316 : | else | ||
317 : | num++; | ||
318 : | } | ||
319 : | |||
320 : | } while ( (oldnum != num) && (num>= minblocks) ); | ||
321 : | |||
322 : | if (num < minblocks) | ||
323 : | { | ||
324 : | const int iEdgedWidth = pParam->edged_width; | ||
325 : | num = 0; | ||
326 : | |||
327 : | /* fprintf(stderr,"Warning! Unreliable GME (%d/%d blocks), falling back to translation.\n",num,MBh*MBw); | ||
328 : | */ | ||
329 : | gmc.duv[0].x= gmc.duv[0].y= gmc.duv[1].x= gmc.duv[1].y= gmc.duv[2].x= gmc.duv[2].y=0; | ||
330 : | |||
331 : | if (!(current->motion_flags & XVID_ME_GME_REFINE)) | ||
332 : | return gmc; | ||
333 : | |||
334 : | for (my = 1; my < (uint32_t)MBh-1; my++) /* ignore boundary blocks */ | ||
335 : | for (mx = 1; mx < (uint32_t)MBw-1; mx++) /* theirs MVs are often wrong */ | ||
336 : | { | ||
337 : | const int mbnum = mx + my * MBw; | ||
338 : | MACROBLOCK *const pMB = &pMBs[mbnum]; | ||
339 : | const uint8_t *const pCur = current->image.y + 16*(my*iEdgedWidth + mx); | ||
340 : | if ( (sad16 ( pCur, pCur+1 , iEdgedWidth, 65536) >= gradx ) | ||
341 : | && (sad16 ( pCur, pCur+iEdgedWidth, iEdgedWidth, 65536) >= grady ) ) | ||
342 : | { pMB->mcsel = 1; | ||
343 : | gmc.duv[0].x += pMB->mvs[0].x; | ||
344 : | gmc.duv[0].y += pMB->mvs[0].y; | ||
345 : | num++; | ||
346 : | } | ||
347 : | } | ||
348 : | |||
349 : | if (gmc.duv[0].x) | ||
350 : | gmc.duv[0].x /= num; | ||
351 : | if (gmc.duv[0].y) | ||
352 : | gmc.duv[0].y /= num; | ||
353 : | } else { | ||
354 : | |||
355 : | gmc.duv[0].x=(int)(sol[0]+0.5); | ||
356 : | gmc.duv[0].y=(int)(sol[3]+0.5); | ||
357 : | |||
358 : | gmc.duv[1].x=(int)(sol[1]*pParam->width+0.5); | ||
359 : | gmc.duv[1].y=(int)(-sol[2]*pParam->width+0.5); | ||
360 : | |||
361 : | gmc.duv[2].x=-gmc.duv[1].y; /* two warp points only */ | ||
362 : | gmc.duv[2].y=gmc.duv[1].x; | ||
363 : | } | ||
364 : | if (num>maxblocks) | ||
365 : | { for (my = 1; my < (uint32_t)MBh-1; my++) | ||
366 : | for (mx = 1; mx < (uint32_t)MBw-1; mx++) | ||
367 : | { | ||
368 : | const int mbnum = mx + my * MBw; | ||
369 : | if (pMBs[mbnum-1].mcsel) | ||
370 : | pMBs[mbnum].mcsel=0; | ||
371 : | else | ||
372 : | if (pMBs[mbnum-MBw].mcsel) | ||
373 : | pMBs[mbnum].mcsel=0; | ||
374 : | } | ||
375 : | } | ||
376 : | return gmc; | ||
377 : | } | ||
378 : | |||
379 : | int | ||
380 : | GlobalMotionEstRefine( | ||
381 : | WARPPOINTS *const startwp, | ||
382 : | MACROBLOCK * const pMBs, | ||
383 : | const MBParam * const pParam, | ||
384 : | const FRAMEINFO * const current, | ||
385 : | const FRAMEINFO * const reference, | ||
386 : | const IMAGE * const pCurr, | ||
387 : | const IMAGE * const pRef, | ||
388 : | const IMAGE * const pRefH, | ||
389 : | const IMAGE * const pRefV, | ||
390 : | const IMAGE * const pRefHV) | ||
391 : | { | ||
392 : | uint8_t* GMCblock = (uint8_t*)malloc(16*pParam->edged_width); | ||
393 : | WARPPOINTS bestwp=*startwp; | ||
394 : | WARPPOINTS centerwp,currwp; | ||
395 : | int gmcminSAD=0; | ||
396 : | int gmcSAD=0; | ||
397 : | int direction; | ||
398 : | #if 0 | ||
399 : | int mx,my; | ||
400 : | #endif | ||
401 : | |||
402 : | #if 0 | ||
403 : | /* use many blocks... */ | ||
404 : | for (my = 0; my < (uint32_t)pParam->mb_height; my++) { | ||
405 : | for (mx = 0; mx < (uint32_t)pParam->mb_width; mx++) { | ||
406 : | const int mbnum = mx + my * pParam->mb_width; | ||
407 : | pMBs[mbnum].mcsel=1; | ||
408 : | } | ||
409 : | } | ||
410 : | #endif | ||
411 : | |||
412 : | #if 0 | ||
413 : | /* or rather don't use too many blocks... */ | ||
414 : | for (my = 1; my < (uint32_t)MBh-1; my++) { | ||
415 : | for (mx = 1; mx < (uint32_t)MBw-1; mx++) { | ||
416 : | const int mbnum = mx + my * MBw; | ||
417 : | if (MBmask[mbnum-1]) | ||
418 : | MBmask[mbnum-1]=0; | ||
419 : | else | ||
420 : | if (MBmask[mbnum-MBw]) | ||
421 : | MBmask[mbnum-1]=0; | ||
422 : | |||
423 : | } | ||
424 : | } | ||
425 : | #endif | ||
426 : | |||
427 : | gmcminSAD = globalSAD(&bestwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
428 : | |||
429 : | if ( (reference->coding_type == S_VOP) | ||
430 : | && ( (reference->warp.duv[1].x != bestwp.duv[1].x) | ||
431 : | || (reference->warp.duv[1].y != bestwp.duv[1].y) | ||
432 : | || (reference->warp.duv[0].x != bestwp.duv[0].x) | ||
433 : | || (reference->warp.duv[0].y != bestwp.duv[0].y) | ||
434 : | || (reference->warp.duv[2].x != bestwp.duv[2].x) | ||
435 : | || (reference->warp.duv[2].y != bestwp.duv[2].y) ) ) | ||
436 : | { | ||
437 : | gmcSAD = globalSAD(&reference->warp, pParam, pMBs, | ||
438 : | current, pRef, pCurr, GMCblock); | ||
439 : | |||
440 : | if (gmcSAD < gmcminSAD) | ||
441 : | { bestwp = reference->warp; | ||
442 : | gmcminSAD = gmcSAD; | ||
443 : | } | ||
444 : | } | ||
445 : | |||
446 : | do { | ||
447 : | direction = 0; | ||
448 : | centerwp = bestwp; | ||
449 : | |||
450 : | currwp = centerwp; | ||
451 : | |||
452 : | currwp.duv[0].x--; | ||
453 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
454 : | if (gmcSAD < gmcminSAD) | ||
455 : | { bestwp = currwp; | ||
456 : | gmcminSAD = gmcSAD; | ||
457 : | direction = 1; | ||
458 : | } | ||
459 : | else | ||
460 : | { | ||
461 : | currwp = centerwp; currwp.duv[0].x++; | ||
462 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
463 : | if (gmcSAD < gmcminSAD) | ||
464 : | { bestwp = currwp; | ||
465 : | gmcminSAD = gmcSAD; | ||
466 : | direction = 2; | ||
467 : | } | ||
468 : | } | ||
469 : | if (direction) continue; | ||
470 : | |||
471 : | currwp = centerwp; currwp.duv[0].y--; | ||
472 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
473 : | if (gmcSAD < gmcminSAD) | ||
474 : | { bestwp = currwp; | ||
475 : | gmcminSAD = gmcSAD; | ||
476 : | direction = 4; | ||
477 : | } | ||
478 : | else | ||
479 : | { | ||
480 : | currwp = centerwp; currwp.duv[0].y++; | ||
481 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
482 : | if (gmcSAD < gmcminSAD) | ||
483 : | { bestwp = currwp; | ||
484 : | gmcminSAD = gmcSAD; | ||
485 : | direction = 8; | ||
486 : | } | ||
487 : | } | ||
488 : | if (direction) continue; | ||
489 : | |||
490 : | currwp = centerwp; currwp.duv[1].x++; | ||
491 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
492 : | if (gmcSAD < gmcminSAD) | ||
493 : | { bestwp = currwp; | ||
494 : | gmcminSAD = gmcSAD; | ||
495 : | direction = 32; | ||
496 : | } | ||
497 : | currwp.duv[2].y++; | ||
498 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
499 : | if (gmcSAD < gmcminSAD) | ||
500 : | { bestwp = currwp; | ||
501 : | gmcminSAD = gmcSAD; | ||
502 : | direction = 1024; | ||
503 : | } | ||
504 : | |||
505 : | currwp = centerwp; currwp.duv[1].x--; | ||
506 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
507 : | if (gmcSAD < gmcminSAD) | ||
508 : | { bestwp = currwp; | ||
509 : | gmcminSAD = gmcSAD; | ||
510 : | direction = 16; | ||
511 : | } | ||
512 : | else | ||
513 : | { | ||
514 : | currwp = centerwp; currwp.duv[1].x++; | ||
515 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
516 : | if (gmcSAD < gmcminSAD) | ||
517 : | { bestwp = currwp; | ||
518 : | gmcminSAD = gmcSAD; | ||
519 : | direction = 32; | ||
520 : | } | ||
521 : | } | ||
522 : | if (direction) continue; | ||
523 : | |||
524 : | |||
525 : | currwp = centerwp; currwp.duv[1].y--; | ||
526 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
527 : | if (gmcSAD < gmcminSAD) | ||
528 : | { bestwp = currwp; | ||
529 : | gmcminSAD = gmcSAD; | ||
530 : | direction = 64; | ||
531 : | } | ||
532 : | else | ||
533 : | { | ||
534 : | currwp = centerwp; currwp.duv[1].y++; | ||
535 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
536 : | if (gmcSAD < gmcminSAD) | ||
537 : | { bestwp = currwp; | ||
538 : | gmcminSAD = gmcSAD; | ||
539 : | direction = 128; | ||
540 : | } | ||
541 : | } | ||
542 : | if (direction) continue; | ||
543 : | |||
544 : | currwp = centerwp; currwp.duv[2].x--; | ||
545 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
546 : | if (gmcSAD < gmcminSAD) | ||
547 : | { bestwp = currwp; | ||
548 : | gmcminSAD = gmcSAD; | ||
549 : | direction = 256; | ||
550 : | } | ||
551 : | else | ||
552 : | { | ||
553 : | currwp = centerwp; currwp.duv[2].x++; | ||
554 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
555 : | if (gmcSAD < gmcminSAD) | ||
556 : | { bestwp = currwp; | ||
557 : | gmcminSAD = gmcSAD; | ||
558 : | direction = 512; | ||
559 : | } | ||
560 : | } | ||
561 : | if (direction) continue; | ||
562 : | |||
563 : | currwp = centerwp; currwp.duv[2].y--; | ||
564 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
565 : | if (gmcSAD < gmcminSAD) | ||
566 : | { bestwp = currwp; | ||
567 : | gmcminSAD = gmcSAD; | ||
568 : | direction = 1024; | ||
569 : | } | ||
570 : | else | ||
571 : | { | ||
572 : | currwp = centerwp; currwp.duv[2].y++; | ||
573 : | gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); | ||
574 : | if (gmcSAD < gmcminSAD) | ||
575 : | { bestwp = currwp; | ||
576 : | gmcminSAD = gmcSAD; | ||
577 : | direction = 2048; | ||
578 : | } | ||
579 : | } | ||
580 : | } while (direction); | ||
581 : | free(GMCblock); | ||
582 : | |||
583 : | *startwp = bestwp; | ||
584 : | |||
585 : | return gmcminSAD; | ||
586 : | } | ||
587 : | |||
588 : | int | ||
589 : | globalSAD(const WARPPOINTS *const wp, | ||
590 : | const MBParam * const pParam, | ||
591 : | const MACROBLOCK * const pMBs, | ||
592 : | const FRAMEINFO * const current, | ||
593 : | const IMAGE * const pRef, | ||
594 : | const IMAGE * const pCurr, | ||
595 : | uint8_t *const GMCblock) | ||
596 : | { | ||
597 : | NEW_GMC_DATA gmc_data; | ||
598 : | int iSAD, gmcSAD=0; | ||
599 : | int num=0; | ||
600 : | unsigned int mx, my; | ||
601 : | |||
602 : | generate_GMCparameters( 3, 3, wp, pParam->width, pParam->height, &gmc_data); | ||
603 : | |||
604 : | for (my = 0; my < (uint32_t)pParam->mb_height; my++) | ||
605 : | for (mx = 0; mx < (uint32_t)pParam->mb_width; mx++) { | ||
606 : | |||
607 : | const int mbnum = mx + my * pParam->mb_width; | ||
608 : | const int iEdgedWidth = pParam->edged_width; | ||
609 : | |||
610 : | if (!pMBs[mbnum].mcsel) | ||
611 : | continue; | ||
612 : | |||
613 : | gmc_data.predict_16x16(&gmc_data, GMCblock, | ||
614 : | pRef->y, | ||
615 : | iEdgedWidth, | ||
616 : | iEdgedWidth, | ||
617 : | mx, my, | ||
618 : | pParam->m_rounding_type); | ||
619 : | |||
620 : | iSAD = sad16 ( pCurr->y + 16*(my*iEdgedWidth + mx), | ||
621 : | GMCblock , iEdgedWidth, 65536); | ||
622 : | iSAD -= pMBs[mbnum].sad16; | ||
623 : | |||
624 : | if (iSAD<0) | ||
625 : | gmcSAD += iSAD; | ||
626 : | num++; | ||
627 : | } | ||
628 : | return gmcSAD; | ||
629 : | } |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |