Parent Directory
|
Revision Log
Revision 1074 - (view) (download)
1 : | edgomez | 851 | /************************************************************************** |
2 : | * | ||
3 : | * XVID MPEG-4 VIDEO CODEC | ||
4 : | * - Motion estimation header - | ||
5 : | * | ||
6 : | * This program is an implementation of a part of one or more MPEG-4 | ||
7 : | * Video tools as specified in ISO/IEC 14496-2 standard. Those intending | ||
8 : | * to use this software module in hardware or software products are | ||
9 : | * advised that its use may infringe existing patents or copyrights, and | ||
10 : | * any such use would be at such party's own risk. The original | ||
11 : | * developer of this software module and his/her company, and subsequent | ||
12 : | * editors and their companies, will have no liability for use of this | ||
13 : | * software or modifications or derivatives thereof. | ||
14 : | * | ||
15 : | * This program is free software; you can redistribute it and/or modify | ||
16 : | * it under the terms of the GNU General Public License as published by | ||
17 : | * the Free Software Foundation; either version 2 of the License, or | ||
18 : | * (at your option) any later version. | ||
19 : | * | ||
20 : | * This program is distributed in the hope that it will be useful, | ||
21 : | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
22 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
23 : | * GNU General Public License for more details. | ||
24 : | * | ||
25 : | * You should have received a copy of the GNU General Public License | ||
26 : | * along with this program; if not, write to the Free Software | ||
27 : | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
28 : | * | ||
29 : | syskin | 1074 | * $Id: motion_est.h,v 1.10 2003-06-26 10:37:42 syskin Exp $ |
30 : | edgomez | 851 | * |
31 : | ***************************************************************************/ | ||
32 : | |||
33 : | #ifndef _MOTION_EST_H_ | ||
34 : | #define _MOTION_EST_H_ | ||
35 : | |||
36 : | #include "../portab.h" | ||
37 : | #include "../global.h" | ||
38 : | #include "../image/reduced.h" | ||
39 : | |||
40 : | /* hard coded motion search parameters for motion_est and smp_motion_est */ | ||
41 : | |||
42 : | // very large value | ||
43 : | #define MV_MAX_ERROR (4096 * 256) | ||
44 : | |||
45 : | /* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ | ||
46 : | #define MV16_INTER_BIAS 512 | ||
47 : | |||
48 : | /* vector map (vlc delta size) smoother parameters ! float !*/ | ||
49 : | #define NEIGH_TEND_16X16 10.5 | ||
50 : | #define NEIGH_TEND_8X8 40.0 | ||
51 : | #define NEIGH_8X8_BIAS 30 | ||
52 : | |||
53 : | syskin | 1021 | #define BITS_MULT 16 |
54 : | |||
55 : | edgomez | 851 | /* Parameters which control inter/inter4v decision */ |
56 : | #define IMV16X16 2 | ||
57 : | |||
58 : | static const int lambda_vec16[32] = | ||
59 : | { 0 ,(int)(1.00235 * NEIGH_TEND_16X16 + 0.5), | ||
60 : | (int)(1.15582*NEIGH_TEND_16X16 + 0.5), (int)(1.31976*NEIGH_TEND_16X16 + 0.5), | ||
61 : | (int)(1.49591*NEIGH_TEND_16X16 + 0.5), (int)(1.68601*NEIGH_TEND_16X16 + 0.5), | ||
62 : | (int)(1.89187*NEIGH_TEND_16X16 + 0.5), (int)(2.11542*NEIGH_TEND_16X16 + 0.5), | ||
63 : | (int)(2.35878*NEIGH_TEND_16X16 + 0.5), (int)(2.62429*NEIGH_TEND_16X16 + 0.5), | ||
64 : | (int)(2.91455*NEIGH_TEND_16X16 + 0.5), (int)(3.23253*NEIGH_TEND_16X16 + 0.5), | ||
65 : | (int)(3.58158*NEIGH_TEND_16X16 + 0.5), (int)(3.96555*NEIGH_TEND_16X16 + 0.5), | ||
66 : | (int)(4.38887*NEIGH_TEND_16X16 + 0.5), (int)(4.85673*NEIGH_TEND_16X16 + 0.5), | ||
67 : | (int)(5.37519*NEIGH_TEND_16X16 + 0.5), (int)(5.95144*NEIGH_TEND_16X16 + 0.5), | ||
68 : | (int)(6.59408*NEIGH_TEND_16X16 + 0.5), (int)(7.31349*NEIGH_TEND_16X16 + 0.5), | ||
69 : | (int)(8.12242*NEIGH_TEND_16X16 + 0.5), (int)(9.03669*NEIGH_TEND_16X16 + 0.5), | ||
70 : | (int)(10.0763*NEIGH_TEND_16X16 + 0.5), (int)(11.2669*NEIGH_TEND_16X16 + 0.5), | ||
71 : | (int)(12.6426*NEIGH_TEND_16X16 + 0.5), (int)(14.2493*NEIGH_TEND_16X16 + 0.5), | ||
72 : | (int)(16.1512*NEIGH_TEND_16X16 + 0.5), (int)(18.442*NEIGH_TEND_16X16 + 0.5), | ||
73 : | (int)(21.2656*NEIGH_TEND_16X16 + 0.5), (int)(24.8580*NEIGH_TEND_16X16 + 0.5), | ||
74 : | (int)(29.6436*NEIGH_TEND_16X16 + 0.5), (int)(36.4949*NEIGH_TEND_16X16 + 0.5) }; | ||
75 : | |||
76 : | static const int lambda_vec8[32] = | ||
77 : | { 0 ,(int)(1.00235 * NEIGH_TEND_8X8 + 0.5), | ||
78 : | (int)(1.15582 + NEIGH_TEND_8X8 + 0.5), (int)(1.31976*NEIGH_TEND_8X8 + 0.5), | ||
79 : | (int)(1.49591*NEIGH_TEND_8X8 + 0.5), (int)(1.68601*NEIGH_TEND_8X8 + 0.5), | ||
80 : | (int)(1.89187*NEIGH_TEND_8X8 + 0.5), (int)(2.11542*NEIGH_TEND_8X8 + 0.5), | ||
81 : | (int)(2.35878*NEIGH_TEND_8X8 + 0.5), (int)(2.62429*NEIGH_TEND_8X8 + 0.5), | ||
82 : | (int)(2.91455*NEIGH_TEND_8X8 + 0.5), (int)(3.23253*NEIGH_TEND_8X8 + 0.5), | ||
83 : | (int)(3.58158*NEIGH_TEND_8X8 + 0.5), (int)(3.96555*NEIGH_TEND_8X8 + 0.5), | ||
84 : | (int)(4.38887*NEIGH_TEND_8X8 + 0.5), (int)(4.85673*NEIGH_TEND_8X8 + 0.5), | ||
85 : | (int)(5.37519*NEIGH_TEND_8X8 + 0.5), (int)(5.95144*NEIGH_TEND_8X8 + 0.5), | ||
86 : | (int)(6.59408*NEIGH_TEND_8X8 + 0.5), (int)(7.31349*NEIGH_TEND_8X8 + 0.5), | ||
87 : | (int)(8.12242*NEIGH_TEND_8X8 + 0.5), (int)(9.03669*NEIGH_TEND_8X8 + 0.5), | ||
88 : | (int)(10.0763*NEIGH_TEND_8X8 + 0.5), (int)(11.2669*NEIGH_TEND_8X8 + 0.5), | ||
89 : | (int)(12.6426*NEIGH_TEND_8X8 + 0.5), (int)(14.2493*NEIGH_TEND_8X8 + 0.5), | ||
90 : | (int)(16.1512*NEIGH_TEND_8X8 + 0.5), (int)(18.442*NEIGH_TEND_8X8 + 0.5), | ||
91 : | (int)(21.2656*NEIGH_TEND_8X8 + 0.5), (int)(24.8580*NEIGH_TEND_8X8 + 0.5), | ||
92 : | (int)(29.6436*NEIGH_TEND_8X8 + 0.5), (int)(36.4949*NEIGH_TEND_8X8 + 0.5) }; | ||
93 : | |||
94 : | // mv.length table | ||
95 : | syskin | 973 | static const int mvtab[64] = { |
96 : | 1, 2, 3, 4, 6, 7, 7, 7, | ||
97 : | 9, 9, 9, 10, 10, 10, 10, 10, | ||
98 : | 10, 10, 10, 10, 10, 10, 10, 10, | ||
99 : | 10, 11, 11, 11, 11, 11, 11, 12, | ||
100 : | 12, 12, 12, 12, 12, 12, 12, 12, | ||
101 : | 12, 12, 12, 12, 12, 12, 12, 12, | ||
102 : | 12, 12, 12, 12, 12, 12, 12, 12, 12 }; | ||
103 : | edgomez | 851 | |
104 : | static const int DQtab[4] = { | ||
105 : | -1, -2, 1, 2 | ||
106 : | }; | ||
107 : | |||
108 : | #define RRV_MV_SCALEDOWN(a) ( (a)>=0 ? (a+1)/2 : (a-1)/2 ) | ||
109 : | |||
110 : | typedef struct | ||
111 : | { | ||
112 : | // general fields | ||
113 : | int max_dx, min_dx, max_dy, min_dy; | ||
114 : | uint32_t rounding; | ||
115 : | VECTOR predMV; | ||
116 : | VECTOR * currentMV; | ||
117 : | VECTOR * currentQMV; | ||
118 : | int32_t * iMinSAD; | ||
119 : | syskin | 976 | const uint8_t * RefP[6]; // N, V, H, HV, cU, cV |
120 : | edgomez | 851 | const uint8_t * CurU; |
121 : | const uint8_t * CurV; | ||
122 : | uint8_t * RefQ; | ||
123 : | const uint8_t * Cur; | ||
124 : | uint32_t lambda16; | ||
125 : | uint32_t lambda8; | ||
126 : | uint32_t iEdgedWidth; | ||
127 : | uint32_t iFcode; | ||
128 : | int * temp; | ||
129 : | int qpel, qpel_precision; | ||
130 : | int chroma; | ||
131 : | int rrv; | ||
132 : | //fields for interpolate and direct modes | ||
133 : | syskin | 976 | const uint8_t * b_RefP[6]; // N, V, H, HV, cU, cV |
134 : | edgomez | 851 | VECTOR bpredMV; |
135 : | uint32_t bFcode; | ||
136 : | // fields for direct mode | ||
137 : | VECTOR directmvF[4]; | ||
138 : | VECTOR directmvB[4]; | ||
139 : | const VECTOR * referencemv; | ||
140 : | syskin | 1021 | // BITS/R-D stuff |
141 : | edgomez | 960 | int16_t * dctSpace; |
142 : | syskin | 1016 | uint32_t iQuant; |
143 : | uint32_t quant_type; | ||
144 : | edgomez | 851 | |
145 : | } SearchData; | ||
146 : | |||
147 : | |||
148 : | typedef void(CheckFunc)(const int x, const int y, | ||
149 : | const int Direction, int * const dir, | ||
150 : | const SearchData * const Data); | ||
151 : | CheckFunc *CheckCandidate; | ||
152 : | |||
153 : | /* | ||
154 : | * Calculate the min/max range | ||
155 : | * relative to the _MACROBLOCK_ position | ||
156 : | */ | ||
157 : | static void __inline | ||
158 : | get_range(int32_t * const min_dx, | ||
159 : | int32_t * const max_dx, | ||
160 : | int32_t * const min_dy, | ||
161 : | int32_t * const max_dy, | ||
162 : | const uint32_t x, | ||
163 : | const uint32_t y, | ||
164 : | uint32_t block_sz, /* block dimension, 8 or 16 */ | ||
165 : | const uint32_t width, | ||
166 : | const uint32_t height, | ||
167 : | const uint32_t fcode, | ||
168 : | const int qpel, /* 1 if the resulting range should be in qpel precision; otherwise 0 */ | ||
169 : | const int rrv) | ||
170 : | { | ||
171 : | int k, m = qpel ? 4 : 2; | ||
172 : | const int search_range = 32 << (fcode - 1); | ||
173 : | int high = search_range - 1; | ||
174 : | int low = -search_range; | ||
175 : | |||
176 : | if (rrv) { | ||
177 : | high = RRV_MV_SCALEUP(high); | ||
178 : | low = RRV_MV_SCALEUP(low); | ||
179 : | block_sz *= 2; | ||
180 : | } | ||
181 : | |||
182 : | k = m * (int)(width - x * block_sz); | ||
183 : | *max_dx = MIN(high, k); | ||
184 : | k = m * (int)(height - y * block_sz); | ||
185 : | *max_dy = MIN(high, k); | ||
186 : | |||
187 : | k = -m * (int)((x+1) * block_sz); | ||
188 : | *min_dx = MAX(low, k); | ||
189 : | k = -m * (int)((y+1) * block_sz); | ||
190 : | *min_dy = MAX(low, k); | ||
191 : | } | ||
192 : | |||
193 : | typedef void MainSearchFunc(int x, int y, const SearchData * const Data, int bDirection); | ||
194 : | |||
195 : | static MainSearchFunc DiamondSearch, AdvDiamondSearch, SquareSearch; | ||
196 : | |||
197 : | static void Search8(const SearchData * const OldData, | ||
198 : | const int x, const int y, | ||
199 : | const uint32_t MotionFlags, | ||
200 : | const MBParam * const pParam, | ||
201 : | MACROBLOCK * const pMB, | ||
202 : | const MACROBLOCK * const pMBs, | ||
203 : | const int block, | ||
204 : | SearchData * const Data); | ||
205 : | |||
206 : | bool | ||
207 : | MotionEstimation(MBParam * const pParam, | ||
208 : | syskin | 884 | FRAMEINFO * const current, |
209 : | FRAMEINFO * const reference, | ||
210 : | const IMAGE * const pRefH, | ||
211 : | const IMAGE * const pRefV, | ||
212 : | const IMAGE * const pRefHV, | ||
213 : | const uint32_t iLimit); | ||
214 : | edgomez | 851 | |
215 : | static void | ||
216 : | SearchP(const IMAGE * const pRef, | ||
217 : | const uint8_t * const pRefH, | ||
218 : | const uint8_t * const pRefV, | ||
219 : | const uint8_t * const pRefHV, | ||
220 : | const IMAGE * const pCur, | ||
221 : | const int x, | ||
222 : | const int y, | ||
223 : | const uint32_t MotionFlags, | ||
224 : | const uint32_t GlobalFlags, | ||
225 : | SearchData * const Data, | ||
226 : | const MBParam * const pParam, | ||
227 : | const MACROBLOCK * const pMBs, | ||
228 : | const MACROBLOCK * const prevMBs, | ||
229 : | MACROBLOCK * const pMB); | ||
230 : | |||
231 : | |||
232 : | static WARPPOINTS | ||
233 : | GlobalMotionEst(const MACROBLOCK * const pMBs, | ||
234 : | const MBParam * const pParam, | ||
235 : | const FRAMEINFO * const current, | ||
236 : | const FRAMEINFO * const reference, | ||
237 : | const IMAGE * const pRefH, | ||
238 : | const IMAGE * const pRefV, | ||
239 : | const IMAGE * const pRefHV ); | ||
240 : | |||
241 : | #define iDiamondSize 2 | ||
242 : | |||
243 : | static __inline uint32_t | ||
244 : | MakeGoodMotionFlags(const uint32_t MotionFlags, const uint32_t GlobalFlags) | ||
245 : | { | ||
246 : | uint32_t Flags = MotionFlags; | ||
247 : | |||
248 : | if (!(GlobalFlags & XVID_MODEDECISION_BITS)) | ||
249 : | Flags &= ~(QUARTERPELREFINE16_BITS+QUARTERPELREFINE8_BITS+HALFPELREFINE16_BITS+HALFPELREFINE8_BITS+EXTSEARCH_BITS); | ||
250 : | |||
251 : | if (Flags & EXTSEARCH_BITS) | ||
252 : | Flags |= HALFPELREFINE16_BITS; | ||
253 : | |||
254 : | if (Flags & EXTSEARCH_BITS && MotionFlags & PMV_EXTSEARCH8) | ||
255 : | Flags |= HALFPELREFINE8_BITS; | ||
256 : | |||
257 : | if (Flags & HALFPELREFINE16_BITS) | ||
258 : | Flags |= QUARTERPELREFINE16_BITS; | ||
259 : | |||
260 : | if (Flags & HALFPELREFINE8_BITS) { | ||
261 : | Flags |= QUARTERPELREFINE8_BITS; | ||
262 : | Flags &= ~PMV_HALFPELREFINE8; | ||
263 : | } | ||
264 : | |||
265 : | if (Flags & QUARTERPELREFINE8_BITS) | ||
266 : | Flags &= ~PMV_QUARTERPELREFINE8; | ||
267 : | |||
268 : | syskin | 1021 | if (Flags & QUARTERPELREFINE16_BITS) |
269 : | Flags &= ~PMV_QUARTERPELREFINE16; | ||
270 : | |||
271 : | edgomez | 851 | if (!(GlobalFlags & XVID_QUARTERPEL)) |
272 : | Flags &= ~(PMV_QUARTERPELREFINE16+PMV_QUARTERPELREFINE8+QUARTERPELREFINE16_BITS+QUARTERPELREFINE8_BITS); | ||
273 : | |||
274 : | if (!(GlobalFlags & XVID_HALFPEL)) | ||
275 : | Flags &= ~(PMV_EXTSEARCH16+PMV_HALFPELREFINE16+PMV_HALFPELREFINE8+HALFPELREFINE16_BITS+HALFPELREFINE8_BITS); | ||
276 : | |||
277 : | if (GlobalFlags & (XVID_GREYSCALE + XVID_REDUCED)) | ||
278 : | Flags &= ~(PMV_CHROMA16 + PMV_CHROMA8); | ||
279 : | |||
280 : | return Flags; | ||
281 : | } | ||
282 : | |||
283 : | /* BITS mode decision and search */ | ||
284 : | |||
285 : | #include "../bitstream/zigzag.h" | ||
286 : | #include "../quant/quant_mpeg4.h" | ||
287 : | #include "../quant/quant_h263.h" | ||
288 : | #include "../bitstream/vlc_codes.h" | ||
289 : | syskin | 1016 | #include "../dct/fdct.h" |
290 : | edgomez | 851 | |
291 : | static int | ||
292 : | CountMBBitsInter(SearchData * const Data, | ||
293 : | const MACROBLOCK * const pMBs, const int x, const int y, | ||
294 : | const MBParam * const pParam, | ||
295 : | const uint32_t MotionFlags); | ||
296 : | |||
297 : | static int | ||
298 : | CountMBBitsInter4v(const SearchData * const Data, | ||
299 : | MACROBLOCK * const pMB, const MACROBLOCK * const pMBs, | ||
300 : | const int x, const int y, | ||
301 : | const MBParam * const pParam, const uint32_t MotionFlags, | ||
302 : | const VECTOR * const backup); | ||
303 : | |||
304 : | static int | ||
305 : | CountMBBitsIntra(const SearchData * const Data); | ||
306 : | |||
307 : | int CodeCoeffIntra_CalcBits(const int16_t qcoeff[64], const uint16_t * zigzag); | ||
308 : | int CodeCoeffInter_CalcBits(const int16_t qcoeff[64], const uint16_t * zigzag); | ||
309 : | |||
310 : | syskin | 1074 | /* one over lambda for R-D mode decision and motion search */ |
311 : | #define LAMBDA ( (int)(BITS_MULT/1.0) ) | ||
312 : | syskin | 1016 | |
313 : | static __inline unsigned int | ||
314 : | Block_CalcBits( int16_t * const coeff, | ||
315 : | int16_t * const data, | ||
316 : | syskin | 1074 | int16_t * const dqcoeff, |
317 : | syskin | 1016 | const uint32_t quant, const int quant_type, |
318 : | uint32_t * cbp, | ||
319 : | syskin | 1021 | const int block) |
320 : | syskin | 1016 | { |
321 : | int sum; | ||
322 : | int bits; | ||
323 : | int distortion = 0; | ||
324 : | syskin | 1021 | int i; |
325 : | syskin | 1016 | |
326 : | fdct(data); | ||
327 : | |||
328 : | if (quant_type == 0) sum = quant_inter(coeff, data, quant); | ||
329 : | else sum = quant4_inter(coeff, data, quant); | ||
330 : | |||
331 : | if (sum > 0) { | ||
332 : | *cbp |= 1 << (5 - block); | ||
333 : | syskin | 1021 | bits = BITS_MULT * CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
334 : | syskin | 1016 | } else bits = 0; |
335 : | |||
336 : | syskin | 1074 | if (quant_type == 0) dequant_inter(dqcoeff, coeff, quant); |
337 : | else dequant4_inter(dqcoeff, coeff, quant); | ||
338 : | syskin | 1016 | |
339 : | syskin | 1021 | for (i = 0; i < 64; i++) { |
340 : | syskin | 1074 | distortion += (data[i] - dqcoeff[i])*(data[i] - dqcoeff[i]); |
341 : | syskin | 1016 | } |
342 : | |||
343 : | syskin | 1074 | bits += (LAMBDA*distortion)/(quant*quant); |
344 : | |||
345 : | syskin | 1016 | return bits; |
346 : | } | ||
347 : | |||
348 : | syskin | 1021 | static __inline unsigned int |
349 : | Block_CalcBitsIntra(int16_t * const coeff, | ||
350 : | int16_t * const data, | ||
351 : | syskin | 1074 | int16_t * const dqcoeff, |
352 : | syskin | 1021 | const uint32_t quant, const int quant_type, |
353 : | uint32_t * cbp, | ||
354 : | const int block, | ||
355 : | int * dcpred) | ||
356 : | { | ||
357 : | int bits, i; | ||
358 : | int distortion = 0; | ||
359 : | syskin | 1072 | uint32_t iDcScaler = get_dc_scaler(quant, block < 4); |
360 : | syskin | 1021 | int b_dc; |
361 : | |||
362 : | fdct(data); | ||
363 : | data[0] -= 1024; | ||
364 : | |||
365 : | syskin | 1074 | if (quant_type == 0) quant_intra(coeff, data, quant, iDcScaler); |
366 : | else quant4_intra(coeff, data, quant, iDcScaler); | ||
367 : | syskin | 1021 | |
368 : | b_dc = coeff[0]; | ||
369 : | if (block < 4) { | ||
370 : | coeff[0] -= *dcpred; | ||
371 : | *dcpred = b_dc; | ||
372 : | } | ||
373 : | |||
374 : | bits = BITS_MULT*CodeCoeffIntra_CalcBits(coeff, scan_tables[0]); | ||
375 : | if (bits != 0) *cbp |= 1 << (5 - block); | ||
376 : | |||
377 : | syskin | 1072 | if (block < 4) bits += BITS_MULT*dcy_tab[coeff[0] + 255].len; |
378 : | else bits += BITS_MULT*dcc_tab[coeff[0] + 255].len; | ||
379 : | |||
380 : | syskin | 1021 | coeff[0] = b_dc; |
381 : | syskin | 1074 | if (quant_type == 0) dequant_intra(dqcoeff, coeff, quant, iDcScaler); |
382 : | else dequant4_intra(dqcoeff, coeff, quant, iDcScaler); | ||
383 : | syskin | 1021 | |
384 : | for (i = 0; i < 64; i++) { | ||
385 : | syskin | 1074 | distortion += (data[i] - dqcoeff[i])*(data[i] - dqcoeff[i]); |
386 : | syskin | 1021 | } |
387 : | |||
388 : | syskin | 1074 | bits += (LAMBDA*distortion)/(quant*quant); |
389 : | syskin | 1021 | |
390 : | return bits; | ||
391 : | } | ||
392 : | |||
393 : | edgomez | 851 | #endif /* _MOTION_EST_H_ */ |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |