1 // Copyright 2011 Google Inc. 2 // 3 // This code is licensed under the same terms as WebM: 4 // Software License Agreement: http://www.webmproject.org/license/software/ 5 // Additional IP Rights Grant: http://www.webmproject.org/license/additional/ 6 // ----------------------------------------------------------------------------- 7 // 8 // Macroblock analysis 9 // 10 // Author: Skal (pascal.massimino (at) gmail.com) 11 12 #include <stdlib.h> 13 #include <string.h> 14 #include <assert.h> 15 16 #include "vp8enci.h" 17 #include "cost.h" 18 19 #if defined(__cplusplus) || defined(c_plusplus) 20 extern "C" { 21 #endif 22 23 #define MAX_ITERS_K_MEANS 6 24 25 static int ClipAlpha(int alpha) { 26 return alpha < 0 ? 0 : alpha > 255 ? 255 : alpha; 27 } 28 29 //----------------------------------------------------------------------------- 30 // Smooth the segment map by replacing isolated block by the majority of its 31 // neighbours. 32 33 static void SmoothSegmentMap(VP8Encoder* const enc) { 34 int n, x, y; 35 const int w = enc->mb_w_; 36 const int h = enc->mb_h_; 37 const int majority_cnt_3_x_3_grid = 5; 38 uint8_t* tmp = (uint8_t*)malloc(w * h * sizeof(uint8_t)); 39 40 if (tmp == NULL) return; 41 for (y = 1; y < h - 1; ++y) { 42 for (x = 1; x < w - 1; ++x) { 43 int cnt[NUM_MB_SEGMENTS] = { 0 }; 44 const VP8MBInfo* const mb = &enc->mb_info_[x + w * y]; 45 int majority_seg = mb->segment_; 46 // Check the 8 neighbouring segment values. 47 cnt[mb[-w - 1].segment_]++; // top-left 48 cnt[mb[-w + 0].segment_]++; // top 49 cnt[mb[-w + 1].segment_]++; // top-right 50 cnt[mb[ - 1].segment_]++; // left 51 cnt[mb[ + 1].segment_]++; // right 52 cnt[mb[ w - 1].segment_]++; // bottom-left 53 cnt[mb[ w + 0].segment_]++; // bottom 54 cnt[mb[ w + 1].segment_]++; // bottom-right 55 for (n = 0; n < NUM_MB_SEGMENTS; ++n) { 56 if (cnt[n] >= majority_cnt_3_x_3_grid) { 57 majority_seg = n; 58 } 59 } 60 tmp[x + y * w] = majority_seg; 61 } 62 } 63 for (y = 1; y < h - 1; ++y) { 64 for (x = 1; x < w - 1; ++x) { 65 VP8MBInfo* const mb = &enc->mb_info_[x + w * y]; 66 mb->segment_ = tmp[x + y * w]; 67 } 68 } 69 free(tmp); 70 } 71 72 //----------------------------------------------------------------------------- 73 // Finalize Segment probability based on the coding tree 74 75 static int GetProba(int a, int b) { 76 int proba; 77 const int total = a + b; 78 if (total == 0) return 255; // that's the default probability. 79 proba = (255 * a + total / 2) / total; 80 return proba; 81 } 82 83 static void SetSegmentProbas(VP8Encoder* const enc) { 84 int p[NUM_MB_SEGMENTS] = { 0 }; 85 int n; 86 87 for (n = 0; n < enc->mb_w_ * enc->mb_h_; ++n) { 88 const VP8MBInfo* const mb = &enc->mb_info_[n]; 89 p[mb->segment_]++; 90 } 91 if (enc->pic_->stats) { 92 for (n = 0; n < NUM_MB_SEGMENTS; ++n) { 93 enc->pic_->stats->segment_size[n] = p[n]; 94 } 95 } 96 if (enc->segment_hdr_.num_segments_ > 1) { 97 uint8_t* const probas = enc->proba_.segments_; 98 probas[0] = GetProba(p[0] + p[1], p[2] + p[3]); 99 probas[1] = GetProba(p[0], p[1]); 100 probas[2] = GetProba(p[2], p[3]); 101 102 enc->segment_hdr_.update_map_ = 103 (probas[0] != 255) || (probas[1] != 255) || (probas[2] != 255); 104 enc->segment_hdr_.size_ = 105 p[0] * (VP8BitCost(0, probas[0]) + VP8BitCost(0, probas[1])) + 106 p[1] * (VP8BitCost(0, probas[0]) + VP8BitCost(1, probas[1])) + 107 p[2] * (VP8BitCost(1, probas[0]) + VP8BitCost(0, probas[2])) + 108 p[3] * (VP8BitCost(1, probas[0]) + VP8BitCost(1, probas[2])); 109 } else { 110 enc->segment_hdr_.update_map_ = 0; 111 enc->segment_hdr_.size_ = 0; 112 } 113 } 114 115 static inline int clip(int v, int m, int M) { 116 return v < m ? m : v > M ? M : v; 117 } 118 119 static void SetSegmentAlphas(VP8Encoder* const enc, 120 const int centers[NUM_MB_SEGMENTS], 121 int mid) { 122 const int nb = enc->segment_hdr_.num_segments_; 123 int min = centers[0], max = centers[0]; 124 int n; 125 126 if (nb > 1) { 127 for (n = 0; n < nb; ++n) { 128 if (min > centers[n]) min = centers[n]; 129 if (max < centers[n]) max = centers[n]; 130 } 131 } 132 if (max == min) max = min + 1; 133 assert(mid <= max && mid >= min); 134 for (n = 0; n < nb; ++n) { 135 const int alpha = 255 * (centers[n] - mid) / (max - min); 136 const int beta = 255 * (centers[n] - min) / (max - min); 137 enc->dqm_[n].alpha_ = clip(alpha, -127, 127); 138 enc->dqm_[n].beta_ = clip(beta, 0, 255); 139 } 140 } 141 142 //----------------------------------------------------------------------------- 143 // Simplified k-Means, to assign Nb segments based on alpha-histogram 144 145 static void AssignSegments(VP8Encoder* const enc, const int alphas[256]) { 146 const int nb = enc->segment_hdr_.num_segments_; 147 int centers[NUM_MB_SEGMENTS]; 148 int weighted_average; 149 int map[256]; 150 int a, n, k; 151 int min_a = 0, max_a = 255, range_a; 152 // 'int' type is ok for histo, and won't overflow 153 int accum[NUM_MB_SEGMENTS], dist_accum[NUM_MB_SEGMENTS]; 154 155 // bracket the input 156 for (n = 0; n < 256 && alphas[n] == 0; ++n) {} 157 min_a = n; 158 for (n = 255; n > min_a && alphas[n] == 0; --n) {} 159 max_a = n; 160 range_a = max_a - min_a; 161 162 // Spread initial centers evenly 163 for (n = 1, k = 0; n < 2 * nb; n += 2) { 164 centers[k++] = min_a + (n * range_a) / (2 * nb); 165 } 166 167 for (k = 0; k < MAX_ITERS_K_MEANS; ++k) { // few iters are enough 168 int total_weight; 169 int displaced; 170 // Reset stats 171 for (n = 0; n < nb; ++n) { 172 accum[n] = 0; 173 dist_accum[n] = 0; 174 } 175 // Assign nearest center for each 'a' 176 n = 0; // track the nearest center for current 'a' 177 for (a = min_a; a <= max_a; ++a) { 178 if (alphas[a]) { 179 while (n < nb - 1 && abs(a - centers[n + 1]) < abs(a - centers[n])) { 180 n++; 181 } 182 map[a] = n; 183 // accumulate contribution into best centroid 184 dist_accum[n] += a * alphas[a]; 185 accum[n] += alphas[a]; 186 } 187 } 188 // All point are classified. Move the centroids to the 189 // center of their respective cloud. 190 displaced = 0; 191 weighted_average = 0; 192 total_weight = 0; 193 for (n = 0; n < nb; ++n) { 194 if (accum[n]) { 195 const int new_center = (dist_accum[n] + accum[n] / 2) / accum[n]; 196 displaced += abs(centers[n] - new_center); 197 centers[n] = new_center; 198 weighted_average += new_center * accum[n]; 199 total_weight += accum[n]; 200 } 201 } 202 weighted_average = (weighted_average + total_weight / 2) / total_weight; 203 if (displaced < 5) break; // no need to keep on looping... 204 } 205 206 // Map each original value to the closest centroid 207 for (n = 0; n < enc->mb_w_ * enc->mb_h_; ++n) { 208 VP8MBInfo* const mb = &enc->mb_info_[n]; 209 const int a = mb->alpha_; 210 mb->segment_ = map[a]; 211 mb->alpha_ = centers[map[a]]; // just for the record. 212 } 213 214 if (nb > 1) { 215 const int smooth = (enc->config_->preprocessing & 1); 216 if (smooth) SmoothSegmentMap(enc); 217 } 218 219 SetSegmentProbas(enc); // Assign final proba 220 SetSegmentAlphas(enc, centers, weighted_average); // pick some alphas. 221 } 222 223 //----------------------------------------------------------------------------- 224 // Macroblock analysis: collect histogram for each mode, deduce the maximal 225 // susceptibility and set best modes for this macroblock. 226 // Segment assignment is done later. 227 228 // Number of modes to inspect for alpha_ evaluation. For high-quality settings, 229 // we don't need to test all the possible modes during the analysis phase. 230 #define MAX_INTRA16_MODE 2 231 #define MAX_INTRA4_MODE 2 232 #define MAX_UV_MODE 2 233 234 static int MBAnalyzeBestIntra16Mode(VP8EncIterator* const it) { 235 const int max_mode = (it->enc_->method_ >= 3) ? MAX_INTRA16_MODE : 4; 236 int mode; 237 int best_alpha = -1; 238 int best_mode = 0; 239 240 VP8MakeLuma16Preds(it); 241 for (mode = 0; mode < max_mode; ++mode) { 242 const int alpha = VP8CollectHistogram(it->yuv_in_ + Y_OFF, 243 it->yuv_p_ + VP8I16ModeOffsets[mode], 244 0, 16); 245 if (alpha > best_alpha) { 246 best_alpha = alpha; 247 best_mode = mode; 248 } 249 } 250 VP8SetIntra16Mode(it, best_mode); 251 return best_alpha; 252 } 253 254 static int MBAnalyzeBestIntra4Mode(VP8EncIterator* const it, 255 int best_alpha) { 256 int modes[16]; 257 const int max_mode = (it->enc_->method_ >= 3) ? MAX_INTRA4_MODE : NUM_BMODES; 258 int i4_alpha = 0; 259 VP8IteratorStartI4(it); 260 do { 261 int mode; 262 int best_mode_alpha = -1; 263 const uint8_t* const src = it->yuv_in_ + Y_OFF + VP8Scan[it->i4_]; 264 265 VP8MakeIntra4Preds(it); 266 for (mode = 0; mode < max_mode; ++mode) { 267 const int alpha = VP8CollectHistogram(src, 268 it->yuv_p_ + VP8I4ModeOffsets[mode], 269 0, 1); 270 if (alpha > best_mode_alpha) { 271 best_mode_alpha = alpha; 272 modes[it->i4_] = mode; 273 } 274 } 275 i4_alpha += best_mode_alpha; 276 // Note: we reuse the original samples for predictors 277 } while (VP8IteratorRotateI4(it, it->yuv_in_ + Y_OFF)); 278 279 if (i4_alpha > best_alpha) { 280 VP8SetIntra4Mode(it, modes); 281 best_alpha = ClipAlpha(i4_alpha); 282 } 283 return best_alpha; 284 } 285 286 static int MBAnalyzeBestUVMode(VP8EncIterator* const it) { 287 int best_alpha = -1; 288 int best_mode = 0; 289 const int max_mode = (it->enc_->method_ >= 3) ? MAX_UV_MODE : 4; 290 int mode; 291 VP8MakeChroma8Preds(it); 292 for (mode = 0; mode < max_mode; ++mode) { 293 const int alpha = VP8CollectHistogram(it->yuv_in_ + U_OFF, 294 it->yuv_p_ + VP8UVModeOffsets[mode], 295 16, 16 + 4 + 4); 296 if (alpha > best_alpha) { 297 best_alpha = alpha; 298 best_mode = mode; 299 } 300 } 301 VP8SetIntraUVMode(it, best_mode); 302 return best_alpha; 303 } 304 305 static void MBAnalyze(VP8EncIterator* const it, 306 int alphas[256], int* const uv_alpha) { 307 const VP8Encoder* const enc = it->enc_; 308 int best_alpha, best_uv_alpha; 309 310 VP8SetIntra16Mode(it, 0); // default: Intra16, DC_PRED 311 VP8SetSkip(it, 0); // not skipped 312 VP8SetSegment(it, 0); // default segment, spec-wise. 313 314 best_alpha = MBAnalyzeBestIntra16Mode(it); 315 if (enc->method_ != 3) { 316 // We go and make a fast decision for intra4/intra16. 317 // It's usually not a good and definitive pick, but helps seeding the stats 318 // about level bit-cost. 319 // TODO(skal): improve criterion. 320 best_alpha = MBAnalyzeBestIntra4Mode(it, best_alpha); 321 } 322 best_uv_alpha = MBAnalyzeBestUVMode(it); 323 324 // Final susceptibility mix 325 best_alpha = (best_alpha + best_uv_alpha + 1) / 2; 326 alphas[best_alpha]++; 327 *uv_alpha += best_uv_alpha; 328 it->mb_->alpha_ = best_alpha; // Informative only. 329 } 330 331 //----------------------------------------------------------------------------- 332 // Main analysis loop: 333 // Collect all susceptibilities for each macroblock and record their 334 // distribution in alphas[]. Segments is assigned a-posteriori, based on 335 // this histogram. 336 // We also pick an intra16 prediction mode, which shouldn't be considered 337 // final except for fast-encode settings. We can also pick some intra4 modes 338 // and decide intra4/intra16, but that's usually almost always a bad choice at 339 // this stage. 340 341 int VP8EncAnalyze(VP8Encoder* const enc) { 342 int alphas[256] = { 0 }; 343 VP8EncIterator it; 344 345 VP8IteratorInit(enc, &it); 346 enc->uv_alpha_ = 0; 347 do { 348 VP8IteratorImport(&it); 349 MBAnalyze(&it, alphas, &enc->uv_alpha_); 350 // Let's pretend we have perfect lossless reconstruction. 351 } while (VP8IteratorNext(&it, it.yuv_in_)); 352 enc->uv_alpha_ /= enc->mb_w_ * enc->mb_h_; 353 AssignSegments(enc, alphas); 354 355 return 1; 356 } 357 358 #if defined(__cplusplus) || defined(c_plusplus) 359 } // extern "C" 360 #endif 361