| 1 | /* |
| 2 | * Copyright (C) 2010 Georg Martius <georg.martius@web.de> |
| 3 | * Copyright (C) 2010 Daniel G. Taylor <dan@programmer-art.org> |
| 4 | * |
| 5 | * This file is part of FFmpeg. |
| 6 | * |
| 7 | * FFmpeg is free software; you can redistribute it and/or |
| 8 | * modify it under the terms of the GNU Lesser General Public |
| 9 | * License as published by the Free Software Foundation; either |
| 10 | * version 2.1 of the License, or (at your option) any later version. |
| 11 | * |
| 12 | * FFmpeg is distributed in the hope that it will be useful, |
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 15 | * Lesser General Public License for more details. |
| 16 | * |
| 17 | * You should have received a copy of the GNU Lesser General Public |
| 18 | * License along with FFmpeg; if not, write to the Free Software |
| 19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 20 | */ |
| 21 | |
| 22 | /** |
| 23 | * @file |
| 24 | * fast deshake / depan video filter |
| 25 | * |
| 26 | * SAD block-matching motion compensation to fix small changes in |
| 27 | * horizontal and/or vertical shift. This filter helps remove camera shake |
| 28 | * from hand-holding a camera, bumping a tripod, moving on a vehicle, etc. |
| 29 | * |
| 30 | * Algorithm: |
| 31 | * - For each frame with one previous reference frame |
| 32 | * - For each block in the frame |
| 33 | * - If contrast > threshold then find likely motion vector |
| 34 | * - For all found motion vectors |
| 35 | * - Find most common, store as global motion vector |
| 36 | * - Find most likely rotation angle |
| 37 | * - Transform image along global motion |
| 38 | * |
| 39 | * TODO: |
| 40 | * - Fill frame edges based on previous/next reference frames |
| 41 | * - Fill frame edges by stretching image near the edges? |
| 42 | * - Can this be done quickly and look decent? |
| 43 | * |
| 44 | * Dark Shikari links to http://wiki.videolan.org/SoC_x264_2010#GPU_Motion_Estimation_2 |
| 45 | * for an algorithm similar to what could be used here to get the gmv |
| 46 | * It requires only a couple diamond searches + fast downscaling |
| 47 | * |
| 48 | * Special thanks to Jason Kotenko for his help with the algorithm and my |
| 49 | * inability to see simple errors in C code. |
| 50 | */ |
| 51 | |
| 52 | #include "avfilter.h" |
| 53 | #include "formats.h" |
| 54 | #include "internal.h" |
| 55 | #include "video.h" |
| 56 | #include "libavutil/common.h" |
| 57 | #include "libavutil/mem.h" |
| 58 | #include "libavutil/opt.h" |
| 59 | #include "libavutil/pixdesc.h" |
| 60 | |
| 61 | #include "deshake.h" |
| 62 | #include "deshake_opencl.h" |
| 63 | |
| 64 | #define CHROMA_WIDTH(link) (-((-(link)->w) >> av_pix_fmt_desc_get((link)->format)->log2_chroma_w)) |
| 65 | #define CHROMA_HEIGHT(link) (-((-(link)->h) >> av_pix_fmt_desc_get((link)->format)->log2_chroma_h)) |
| 66 | |
| 67 | #define OFFSET(x) offsetof(DeshakeContext, x) |
| 68 | #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM |
| 69 | |
| 70 | static const AVOption deshake_options[] = { |
| 71 | { "x", "set x for the rectangular search area", OFFSET(cx), AV_OPT_TYPE_INT, {.i64=-1}, -1, INT_MAX, .flags = FLAGS }, |
| 72 | { "y", "set y for the rectangular search area", OFFSET(cy), AV_OPT_TYPE_INT, {.i64=-1}, -1, INT_MAX, .flags = FLAGS }, |
| 73 | { "w", "set width for the rectangular search area", OFFSET(cw), AV_OPT_TYPE_INT, {.i64=-1}, -1, INT_MAX, .flags = FLAGS }, |
| 74 | { "h", "set height for the rectangular search area", OFFSET(ch), AV_OPT_TYPE_INT, {.i64=-1}, -1, INT_MAX, .flags = FLAGS }, |
| 75 | { "rx", "set x for the rectangular search area", OFFSET(rx), AV_OPT_TYPE_INT, {.i64=16}, 0, MAX_R, .flags = FLAGS }, |
| 76 | { "ry", "set y for the rectangular search area", OFFSET(ry), AV_OPT_TYPE_INT, {.i64=16}, 0, MAX_R, .flags = FLAGS }, |
| 77 | { "edge", "set edge mode", OFFSET(edge), AV_OPT_TYPE_INT, {.i64=FILL_MIRROR}, FILL_BLANK, FILL_COUNT-1, FLAGS, "edge"}, |
| 78 | { "blank", "fill zeroes at blank locations", 0, AV_OPT_TYPE_CONST, {.i64=FILL_BLANK}, INT_MIN, INT_MAX, FLAGS, "edge" }, |
| 79 | { "original", "original image at blank locations", 0, AV_OPT_TYPE_CONST, {.i64=FILL_ORIGINAL}, INT_MIN, INT_MAX, FLAGS, "edge" }, |
| 80 | { "clamp", "extruded edge value at blank locations", 0, AV_OPT_TYPE_CONST, {.i64=FILL_CLAMP}, INT_MIN, INT_MAX, FLAGS, "edge" }, |
| 81 | { "mirror", "mirrored edge at blank locations", 0, AV_OPT_TYPE_CONST, {.i64=FILL_MIRROR}, INT_MIN, INT_MAX, FLAGS, "edge" }, |
| 82 | { "blocksize", "set motion search blocksize", OFFSET(blocksize), AV_OPT_TYPE_INT, {.i64=8}, 4, 128, .flags = FLAGS }, |
| 83 | { "contrast", "set contrast threshold for blocks", OFFSET(contrast), AV_OPT_TYPE_INT, {.i64=125}, 1, 255, .flags = FLAGS }, |
| 84 | { "search", "set search strategy", OFFSET(search), AV_OPT_TYPE_INT, {.i64=EXHAUSTIVE}, EXHAUSTIVE, SEARCH_COUNT-1, FLAGS, "smode" }, |
| 85 | { "exhaustive", "exhaustive search", 0, AV_OPT_TYPE_CONST, {.i64=EXHAUSTIVE}, INT_MIN, INT_MAX, FLAGS, "smode" }, |
| 86 | { "less", "less exhaustive search", 0, AV_OPT_TYPE_CONST, {.i64=SMART_EXHAUSTIVE}, INT_MIN, INT_MAX, FLAGS, "smode" }, |
| 87 | { "filename", "set motion search detailed log file name", OFFSET(filename), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, |
| 88 | { "opencl", "use OpenCL filtering capabilities", OFFSET(opencl), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, .flags = FLAGS }, |
| 89 | { NULL } |
| 90 | }; |
| 91 | |
| 92 | AVFILTER_DEFINE_CLASS(deshake); |
| 93 | |
| 94 | static int cmp(const double *a, const double *b) |
| 95 | { |
| 96 | return *a < *b ? -1 : ( *a > *b ? 1 : 0 ); |
| 97 | } |
| 98 | |
| 99 | /** |
| 100 | * Cleaned mean (cuts off 20% of values to remove outliers and then averages) |
| 101 | */ |
| 102 | static double clean_mean(double *values, int count) |
| 103 | { |
| 104 | double mean = 0; |
| 105 | int cut = count / 5; |
| 106 | int x; |
| 107 | |
| 108 | qsort(values, count, sizeof(double), (void*)cmp); |
| 109 | |
| 110 | for (x = cut; x < count - cut; x++) { |
| 111 | mean += values[x]; |
| 112 | } |
| 113 | |
| 114 | return mean / (count - cut * 2); |
| 115 | } |
| 116 | |
| 117 | /** |
| 118 | * Find the most likely shift in motion between two frames for a given |
| 119 | * macroblock. Test each block against several shifts given by the rx |
| 120 | * and ry attributes. Searches using a simple matrix of those shifts and |
| 121 | * chooses the most likely shift by the smallest difference in blocks. |
| 122 | */ |
| 123 | static void find_block_motion(DeshakeContext *deshake, uint8_t *src1, |
| 124 | uint8_t *src2, int cx, int cy, int stride, |
| 125 | IntMotionVector *mv) |
| 126 | { |
| 127 | int x, y; |
| 128 | int diff; |
| 129 | int smallest = INT_MAX; |
| 130 | int tmp, tmp2; |
| 131 | |
| 132 | #define CMP(i, j) deshake->sad(src1 + cy * stride + cx, stride,\ |
| 133 | src2 + (j) * stride + (i), stride) |
| 134 | |
| 135 | if (deshake->search == EXHAUSTIVE) { |
| 136 | // Compare every possible position - this is sloooow! |
| 137 | for (y = -deshake->ry; y <= deshake->ry; y++) { |
| 138 | for (x = -deshake->rx; x <= deshake->rx; x++) { |
| 139 | diff = CMP(cx - x, cy - y); |
| 140 | if (diff < smallest) { |
| 141 | smallest = diff; |
| 142 | mv->x = x; |
| 143 | mv->y = y; |
| 144 | } |
| 145 | } |
| 146 | } |
| 147 | } else if (deshake->search == SMART_EXHAUSTIVE) { |
| 148 | // Compare every other possible position and find the best match |
| 149 | for (y = -deshake->ry + 1; y < deshake->ry; y += 2) { |
| 150 | for (x = -deshake->rx + 1; x < deshake->rx; x += 2) { |
| 151 | diff = CMP(cx - x, cy - y); |
| 152 | if (diff < smallest) { |
| 153 | smallest = diff; |
| 154 | mv->x = x; |
| 155 | mv->y = y; |
| 156 | } |
| 157 | } |
| 158 | } |
| 159 | |
| 160 | // Hone in on the specific best match around the match we found above |
| 161 | tmp = mv->x; |
| 162 | tmp2 = mv->y; |
| 163 | |
| 164 | for (y = tmp2 - 1; y <= tmp2 + 1; y++) { |
| 165 | for (x = tmp - 1; x <= tmp + 1; x++) { |
| 166 | if (x == tmp && y == tmp2) |
| 167 | continue; |
| 168 | |
| 169 | diff = CMP(cx - x, cy - y); |
| 170 | if (diff < smallest) { |
| 171 | smallest = diff; |
| 172 | mv->x = x; |
| 173 | mv->y = y; |
| 174 | } |
| 175 | } |
| 176 | } |
| 177 | } |
| 178 | |
| 179 | if (smallest > 512) { |
| 180 | mv->x = -1; |
| 181 | mv->y = -1; |
| 182 | } |
| 183 | emms_c(); |
| 184 | //av_log(NULL, AV_LOG_ERROR, "%d\n", smallest); |
| 185 | //av_log(NULL, AV_LOG_ERROR, "Final: (%d, %d) = %d x %d\n", cx, cy, mv->x, mv->y); |
| 186 | } |
| 187 | |
| 188 | /** |
| 189 | * Find the contrast of a given block. When searching for global motion we |
| 190 | * really only care about the high contrast blocks, so using this method we |
| 191 | * can actually skip blocks we don't care much about. |
| 192 | */ |
| 193 | static int block_contrast(uint8_t *src, int x, int y, int stride, int blocksize) |
| 194 | { |
| 195 | int highest = 0; |
| 196 | int lowest = 255; |
| 197 | int i, j, pos; |
| 198 | |
| 199 | for (i = 0; i <= blocksize * 2; i++) { |
| 200 | // We use a width of 16 here to match the sad function |
| 201 | for (j = 0; j <= 15; j++) { |
| 202 | pos = (y - i) * stride + (x - j); |
| 203 | if (src[pos] < lowest) |
| 204 | lowest = src[pos]; |
| 205 | else if (src[pos] > highest) { |
| 206 | highest = src[pos]; |
| 207 | } |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | return highest - lowest; |
| 212 | } |
| 213 | |
| 214 | /** |
| 215 | * Find the rotation for a given block. |
| 216 | */ |
| 217 | static double block_angle(int x, int y, int cx, int cy, IntMotionVector *shift) |
| 218 | { |
| 219 | double a1, a2, diff; |
| 220 | |
| 221 | a1 = atan2(y - cy, x - cx); |
| 222 | a2 = atan2(y - cy + shift->y, x - cx + shift->x); |
| 223 | |
| 224 | diff = a2 - a1; |
| 225 | |
| 226 | return (diff > M_PI) ? diff - 2 * M_PI : |
| 227 | (diff < -M_PI) ? diff + 2 * M_PI : |
| 228 | diff; |
| 229 | } |
| 230 | |
| 231 | /** |
| 232 | * Find the estimated global motion for a scene given the most likely shift |
| 233 | * for each block in the frame. The global motion is estimated to be the |
| 234 | * same as the motion from most blocks in the frame, so if most blocks |
| 235 | * move one pixel to the right and two pixels down, this would yield a |
| 236 | * motion vector (1, -2). |
| 237 | */ |
| 238 | static void find_motion(DeshakeContext *deshake, uint8_t *src1, uint8_t *src2, |
| 239 | int width, int height, int stride, Transform *t) |
| 240 | { |
| 241 | int x, y; |
| 242 | IntMotionVector mv = {0, 0}; |
| 243 | int count_max_value = 0; |
| 244 | int contrast; |
| 245 | |
| 246 | int pos; |
| 247 | int center_x = 0, center_y = 0; |
| 248 | double p_x, p_y; |
| 249 | |
| 250 | av_fast_malloc(&deshake->angles, &deshake->angles_size, width * height / (16 * deshake->blocksize) * sizeof(*deshake->angles)); |
| 251 | |
| 252 | // Reset counts to zero |
| 253 | for (x = 0; x < deshake->rx * 2 + 1; x++) { |
| 254 | for (y = 0; y < deshake->ry * 2 + 1; y++) { |
| 255 | deshake->counts[x][y] = 0; |
| 256 | } |
| 257 | } |
| 258 | |
| 259 | pos = 0; |
| 260 | // Find motion for every block and store the motion vector in the counts |
| 261 | for (y = deshake->ry; y < height - deshake->ry - (deshake->blocksize * 2); y += deshake->blocksize * 2) { |
| 262 | // We use a width of 16 here to match the sad function |
| 263 | for (x = deshake->rx; x < width - deshake->rx - 16; x += 16) { |
| 264 | // If the contrast is too low, just skip this block as it probably |
| 265 | // won't be very useful to us. |
| 266 | contrast = block_contrast(src2, x, y, stride, deshake->blocksize); |
| 267 | if (contrast > deshake->contrast) { |
| 268 | //av_log(NULL, AV_LOG_ERROR, "%d\n", contrast); |
| 269 | find_block_motion(deshake, src1, src2, x, y, stride, &mv); |
| 270 | if (mv.x != -1 && mv.y != -1) { |
| 271 | deshake->counts[mv.x + deshake->rx][mv.y + deshake->ry] += 1; |
| 272 | if (x > deshake->rx && y > deshake->ry) |
| 273 | deshake->angles[pos++] = block_angle(x, y, 0, 0, &mv); |
| 274 | |
| 275 | center_x += mv.x; |
| 276 | center_y += mv.y; |
| 277 | } |
| 278 | } |
| 279 | } |
| 280 | } |
| 281 | |
| 282 | if (pos) { |
| 283 | center_x /= pos; |
| 284 | center_y /= pos; |
| 285 | t->angle = clean_mean(deshake->angles, pos); |
| 286 | if (t->angle < 0.001) |
| 287 | t->angle = 0; |
| 288 | } else { |
| 289 | t->angle = 0; |
| 290 | } |
| 291 | |
| 292 | // Find the most common motion vector in the frame and use it as the gmv |
| 293 | for (y = deshake->ry * 2; y >= 0; y--) { |
| 294 | for (x = 0; x < deshake->rx * 2 + 1; x++) { |
| 295 | //av_log(NULL, AV_LOG_ERROR, "%5d ", deshake->counts[x][y]); |
| 296 | if (deshake->counts[x][y] > count_max_value) { |
| 297 | t->vec.x = x - deshake->rx; |
| 298 | t->vec.y = y - deshake->ry; |
| 299 | count_max_value = deshake->counts[x][y]; |
| 300 | } |
| 301 | } |
| 302 | //av_log(NULL, AV_LOG_ERROR, "\n"); |
| 303 | } |
| 304 | |
| 305 | p_x = (center_x - width / 2.0); |
| 306 | p_y = (center_y - height / 2.0); |
| 307 | t->vec.x += (cos(t->angle)-1)*p_x - sin(t->angle)*p_y; |
| 308 | t->vec.y += sin(t->angle)*p_x + (cos(t->angle)-1)*p_y; |
| 309 | |
| 310 | // Clamp max shift & rotation? |
| 311 | t->vec.x = av_clipf(t->vec.x, -deshake->rx * 2, deshake->rx * 2); |
| 312 | t->vec.y = av_clipf(t->vec.y, -deshake->ry * 2, deshake->ry * 2); |
| 313 | t->angle = av_clipf(t->angle, -0.1, 0.1); |
| 314 | |
| 315 | //av_log(NULL, AV_LOG_ERROR, "%d x %d\n", avg->x, avg->y); |
| 316 | } |
| 317 | |
| 318 | static int deshake_transform_c(AVFilterContext *ctx, |
| 319 | int width, int height, int cw, int ch, |
| 320 | const float *matrix_y, const float *matrix_uv, |
| 321 | enum InterpolateMethod interpolate, |
| 322 | enum FillMethod fill, AVFrame *in, AVFrame *out) |
| 323 | { |
| 324 | int i = 0, ret = 0; |
| 325 | const float *matrixs[3]; |
| 326 | int plane_w[3], plane_h[3]; |
| 327 | matrixs[0] = matrix_y; |
| 328 | matrixs[1] = matrixs[2] = matrix_uv; |
| 329 | plane_w[0] = width; |
| 330 | plane_w[1] = plane_w[2] = cw; |
| 331 | plane_h[0] = height; |
| 332 | plane_h[1] = plane_h[2] = ch; |
| 333 | |
| 334 | for (i = 0; i < 3; i++) { |
| 335 | // Transform the luma and chroma planes |
| 336 | ret = avfilter_transform(in->data[i], out->data[i], in->linesize[i], out->linesize[i], |
| 337 | plane_w[i], plane_h[i], matrixs[i], interpolate, fill); |
| 338 | if (ret < 0) |
| 339 | return ret; |
| 340 | } |
| 341 | return ret; |
| 342 | } |
| 343 | |
| 344 | static av_cold int init(AVFilterContext *ctx) |
| 345 | { |
| 346 | int ret; |
| 347 | DeshakeContext *deshake = ctx->priv; |
| 348 | |
| 349 | deshake->sad = av_pixelutils_get_sad_fn(4, 4, 1, deshake); // 16x16, 2nd source unaligned |
| 350 | if (!deshake->sad) |
| 351 | return AVERROR(EINVAL); |
| 352 | |
| 353 | deshake->refcount = 20; // XXX: add to options? |
| 354 | deshake->blocksize /= 2; |
| 355 | deshake->blocksize = av_clip(deshake->blocksize, 4, 128); |
| 356 | |
| 357 | if (deshake->rx % 16) { |
| 358 | av_log(ctx, AV_LOG_ERROR, "rx must be a multiple of 16\n"); |
| 359 | return AVERROR_PATCHWELCOME; |
| 360 | } |
| 361 | |
| 362 | if (deshake->filename) |
| 363 | deshake->fp = fopen(deshake->filename, "w"); |
| 364 | if (deshake->fp) |
| 365 | fwrite("Ori x, Avg x, Fin x, Ori y, Avg y, Fin y, Ori angle, Avg angle, Fin angle, Ori zoom, Avg zoom, Fin zoom\n", sizeof(char), 104, deshake->fp); |
| 366 | |
| 367 | // Quadword align left edge of box for MMX code, adjust width if necessary |
| 368 | // to keep right margin |
| 369 | if (deshake->cx > 0) { |
| 370 | deshake->cw += deshake->cx - (deshake->cx & ~15); |
| 371 | deshake->cx &= ~15; |
| 372 | } |
| 373 | deshake->transform = deshake_transform_c; |
| 374 | if (!CONFIG_OPENCL && deshake->opencl) { |
| 375 | av_log(ctx, AV_LOG_ERROR, "OpenCL support was not enabled in this build, cannot be selected\n"); |
| 376 | return AVERROR(EINVAL); |
| 377 | } |
| 378 | |
| 379 | if (CONFIG_OPENCL && deshake->opencl) { |
| 380 | deshake->transform = ff_opencl_transform; |
| 381 | ret = ff_opencl_deshake_init(ctx); |
| 382 | if (ret < 0) |
| 383 | return ret; |
| 384 | } |
| 385 | av_log(ctx, AV_LOG_VERBOSE, "cx: %d, cy: %d, cw: %d, ch: %d, rx: %d, ry: %d, edge: %d blocksize: %d contrast: %d search: %d\n", |
| 386 | deshake->cx, deshake->cy, deshake->cw, deshake->ch, |
| 387 | deshake->rx, deshake->ry, deshake->edge, deshake->blocksize * 2, deshake->contrast, deshake->search); |
| 388 | |
| 389 | return 0; |
| 390 | } |
| 391 | |
| 392 | static int query_formats(AVFilterContext *ctx) |
| 393 | { |
| 394 | static const enum AVPixelFormat pix_fmts[] = { |
| 395 | AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV410P, |
| 396 | AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P, |
| 397 | AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_NONE |
| 398 | }; |
| 399 | |
| 400 | ff_set_common_formats(ctx, ff_make_format_list(pix_fmts)); |
| 401 | |
| 402 | return 0; |
| 403 | } |
| 404 | |
| 405 | static int config_props(AVFilterLink *link) |
| 406 | { |
| 407 | DeshakeContext *deshake = link->dst->priv; |
| 408 | |
| 409 | deshake->ref = NULL; |
| 410 | deshake->last.vec.x = 0; |
| 411 | deshake->last.vec.y = 0; |
| 412 | deshake->last.angle = 0; |
| 413 | deshake->last.zoom = 0; |
| 414 | |
| 415 | return 0; |
| 416 | } |
| 417 | |
| 418 | static av_cold void uninit(AVFilterContext *ctx) |
| 419 | { |
| 420 | DeshakeContext *deshake = ctx->priv; |
| 421 | if (CONFIG_OPENCL && deshake->opencl) { |
| 422 | ff_opencl_deshake_uninit(ctx); |
| 423 | } |
| 424 | av_frame_free(&deshake->ref); |
| 425 | av_freep(&deshake->angles); |
| 426 | deshake->angles_size = 0; |
| 427 | if (deshake->fp) |
| 428 | fclose(deshake->fp); |
| 429 | } |
| 430 | |
| 431 | static int filter_frame(AVFilterLink *link, AVFrame *in) |
| 432 | { |
| 433 | DeshakeContext *deshake = link->dst->priv; |
| 434 | AVFilterLink *outlink = link->dst->outputs[0]; |
| 435 | AVFrame *out; |
| 436 | Transform t = {{0},0}, orig = {{0},0}; |
| 437 | float matrix_y[9], matrix_uv[9]; |
| 438 | float alpha = 2.0 / deshake->refcount; |
| 439 | char tmp[256]; |
| 440 | int ret = 0; |
| 441 | |
| 442 | out = ff_get_video_buffer(outlink, outlink->w, outlink->h); |
| 443 | if (!out) { |
| 444 | av_frame_free(&in); |
| 445 | return AVERROR(ENOMEM); |
| 446 | } |
| 447 | av_frame_copy_props(out, in); |
| 448 | |
| 449 | if (CONFIG_OPENCL && deshake->opencl) { |
| 450 | ret = ff_opencl_deshake_process_inout_buf(link->dst,in, out); |
| 451 | if (ret < 0) |
| 452 | return ret; |
| 453 | } |
| 454 | |
| 455 | if (deshake->cx < 0 || deshake->cy < 0 || deshake->cw < 0 || deshake->ch < 0) { |
| 456 | // Find the most likely global motion for the current frame |
| 457 | find_motion(deshake, (deshake->ref == NULL) ? in->data[0] : deshake->ref->data[0], in->data[0], link->w, link->h, in->linesize[0], &t); |
| 458 | } else { |
| 459 | uint8_t *src1 = (deshake->ref == NULL) ? in->data[0] : deshake->ref->data[0]; |
| 460 | uint8_t *src2 = in->data[0]; |
| 461 | |
| 462 | deshake->cx = FFMIN(deshake->cx, link->w); |
| 463 | deshake->cy = FFMIN(deshake->cy, link->h); |
| 464 | |
| 465 | if ((unsigned)deshake->cx + (unsigned)deshake->cw > link->w) deshake->cw = link->w - deshake->cx; |
| 466 | if ((unsigned)deshake->cy + (unsigned)deshake->ch > link->h) deshake->ch = link->h - deshake->cy; |
| 467 | |
| 468 | // Quadword align right margin |
| 469 | deshake->cw &= ~15; |
| 470 | |
| 471 | src1 += deshake->cy * in->linesize[0] + deshake->cx; |
| 472 | src2 += deshake->cy * in->linesize[0] + deshake->cx; |
| 473 | |
| 474 | find_motion(deshake, src1, src2, deshake->cw, deshake->ch, in->linesize[0], &t); |
| 475 | } |
| 476 | |
| 477 | |
| 478 | // Copy transform so we can output it later to compare to the smoothed value |
| 479 | orig.vec.x = t.vec.x; |
| 480 | orig.vec.y = t.vec.y; |
| 481 | orig.angle = t.angle; |
| 482 | orig.zoom = t.zoom; |
| 483 | |
| 484 | // Generate a one-sided moving exponential average |
| 485 | deshake->avg.vec.x = alpha * t.vec.x + (1.0 - alpha) * deshake->avg.vec.x; |
| 486 | deshake->avg.vec.y = alpha * t.vec.y + (1.0 - alpha) * deshake->avg.vec.y; |
| 487 | deshake->avg.angle = alpha * t.angle + (1.0 - alpha) * deshake->avg.angle; |
| 488 | deshake->avg.zoom = alpha * t.zoom + (1.0 - alpha) * deshake->avg.zoom; |
| 489 | |
| 490 | // Remove the average from the current motion to detect the motion that |
| 491 | // is not on purpose, just as jitter from bumping the camera |
| 492 | t.vec.x -= deshake->avg.vec.x; |
| 493 | t.vec.y -= deshake->avg.vec.y; |
| 494 | t.angle -= deshake->avg.angle; |
| 495 | t.zoom -= deshake->avg.zoom; |
| 496 | |
| 497 | // Invert the motion to undo it |
| 498 | t.vec.x *= -1; |
| 499 | t.vec.y *= -1; |
| 500 | t.angle *= -1; |
| 501 | |
| 502 | // Write statistics to file |
| 503 | if (deshake->fp) { |
| 504 | snprintf(tmp, 256, "%f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f\n", orig.vec.x, deshake->avg.vec.x, t.vec.x, orig.vec.y, deshake->avg.vec.y, t.vec.y, orig.angle, deshake->avg.angle, t.angle, orig.zoom, deshake->avg.zoom, t.zoom); |
| 505 | fwrite(tmp, sizeof(char), strlen(tmp), deshake->fp); |
| 506 | } |
| 507 | |
| 508 | // Turn relative current frame motion into absolute by adding it to the |
| 509 | // last absolute motion |
| 510 | t.vec.x += deshake->last.vec.x; |
| 511 | t.vec.y += deshake->last.vec.y; |
| 512 | t.angle += deshake->last.angle; |
| 513 | t.zoom += deshake->last.zoom; |
| 514 | |
| 515 | // Shrink motion by 10% to keep things centered in the camera frame |
| 516 | t.vec.x *= 0.9; |
| 517 | t.vec.y *= 0.9; |
| 518 | t.angle *= 0.9; |
| 519 | |
| 520 | // Store the last absolute motion information |
| 521 | deshake->last.vec.x = t.vec.x; |
| 522 | deshake->last.vec.y = t.vec.y; |
| 523 | deshake->last.angle = t.angle; |
| 524 | deshake->last.zoom = t.zoom; |
| 525 | |
| 526 | // Generate a luma transformation matrix |
| 527 | avfilter_get_matrix(t.vec.x, t.vec.y, t.angle, 1.0 + t.zoom / 100.0, matrix_y); |
| 528 | // Generate a chroma transformation matrix |
| 529 | avfilter_get_matrix(t.vec.x / (link->w / CHROMA_WIDTH(link)), t.vec.y / (link->h / CHROMA_HEIGHT(link)), t.angle, 1.0 + t.zoom / 100.0, matrix_uv); |
| 530 | // Transform the luma and chroma planes |
| 531 | ret = deshake->transform(link->dst, link->w, link->h, CHROMA_WIDTH(link), CHROMA_HEIGHT(link), |
| 532 | matrix_y, matrix_uv, INTERPOLATE_BILINEAR, deshake->edge, in, out); |
| 533 | |
| 534 | // Cleanup the old reference frame |
| 535 | av_frame_free(&deshake->ref); |
| 536 | |
| 537 | if (ret < 0) |
| 538 | return ret; |
| 539 | |
| 540 | // Store the current frame as the reference frame for calculating the |
| 541 | // motion of the next frame |
| 542 | deshake->ref = in; |
| 543 | |
| 544 | return ff_filter_frame(outlink, out); |
| 545 | } |
| 546 | |
| 547 | static const AVFilterPad deshake_inputs[] = { |
| 548 | { |
| 549 | .name = "default", |
| 550 | .type = AVMEDIA_TYPE_VIDEO, |
| 551 | .filter_frame = filter_frame, |
| 552 | .config_props = config_props, |
| 553 | }, |
| 554 | { NULL } |
| 555 | }; |
| 556 | |
| 557 | static const AVFilterPad deshake_outputs[] = { |
| 558 | { |
| 559 | .name = "default", |
| 560 | .type = AVMEDIA_TYPE_VIDEO, |
| 561 | }, |
| 562 | { NULL } |
| 563 | }; |
| 564 | |
| 565 | AVFilter ff_vf_deshake = { |
| 566 | .name = "deshake", |
| 567 | .description = NULL_IF_CONFIG_SMALL("Stabilize shaky video."), |
| 568 | .priv_size = sizeof(DeshakeContext), |
| 569 | .init = init, |
| 570 | .uninit = uninit, |
| 571 | .query_formats = query_formats, |
| 572 | .inputs = deshake_inputs, |
| 573 | .outputs = deshake_outputs, |
| 574 | .priv_class = &deshake_class, |
| 575 | }; |