2 * Copyright (c) 2002 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (c) 2013 Paul B Mahol
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public
9 * License as published by the Free Software Foundation; either
10 * version 2 of the License, or (at your option) any later version.
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
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License along
18 * with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "libavutil/avassert.h"
23 #include "libavutil/eval.h"
24 #include "libavutil/imgutils.h"
25 #include "libavutil/pixdesc.h"
26 #include "libavutil/opt.h"
32 #define SUB_PIXEL_BITS 8
33 #define SUB_PIXELS (1 << SUB_PIXEL_BITS)
39 typedef struct PerspectiveContext {
44 int32_t coeff[SUB_PIXELS][4];
52 int (*perspective)(AVFilterContext *ctx,
53 void *arg, int job, int nb_jobs);
56 #define OFFSET(x) offsetof(PerspectiveContext, x)
57 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
59 enum PERSPECTIVESense {
60 PERSPECTIVE_SENSE_SOURCE = 0, ///< coordinates give locations in source of corners of destination.
61 PERSPECTIVE_SENSE_DESTINATION = 1, ///< coordinates give locations in destination of corners of source.
64 static const AVOption perspective_options[] = {
65 { "x0", "set top left x coordinate", OFFSET(expr_str[0][0]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
66 { "y0", "set top left y coordinate", OFFSET(expr_str[0][1]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
67 { "x1", "set top right x coordinate", OFFSET(expr_str[1][0]), AV_OPT_TYPE_STRING, {.str="W"}, 0, 0, FLAGS },
68 { "y1", "set top right y coordinate", OFFSET(expr_str[1][1]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
69 { "x2", "set bottom left x coordinate", OFFSET(expr_str[2][0]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
70 { "y2", "set bottom left y coordinate", OFFSET(expr_str[2][1]), AV_OPT_TYPE_STRING, {.str="H"}, 0, 0, FLAGS },
71 { "x3", "set bottom right x coordinate", OFFSET(expr_str[3][0]), AV_OPT_TYPE_STRING, {.str="W"}, 0, 0, FLAGS },
72 { "y3", "set bottom right y coordinate", OFFSET(expr_str[3][1]), AV_OPT_TYPE_STRING, {.str="H"}, 0, 0, FLAGS },
73 { "interpolation", "set interpolation", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=LINEAR}, 0, 1, FLAGS, "interpolation" },
74 { "linear", "", 0, AV_OPT_TYPE_CONST, {.i64=LINEAR}, 0, 0, FLAGS, "interpolation" },
75 { "cubic", "", 0, AV_OPT_TYPE_CONST, {.i64=CUBIC}, 0, 0, FLAGS, "interpolation" },
76 { "sense", "specify the sense of the coordinates", OFFSET(sense), AV_OPT_TYPE_INT, {.i64=PERSPECTIVE_SENSE_SOURCE}, 0, 1, FLAGS, "sense"},
77 { "source", "specify locations in source to send to corners in destination",
78 0, AV_OPT_TYPE_CONST, {.i64=PERSPECTIVE_SENSE_SOURCE}, 0, 0, FLAGS, "sense"},
79 { "destination", "specify locations in destination to send corners of source",
80 0, AV_OPT_TYPE_CONST, {.i64=PERSPECTIVE_SENSE_DESTINATION}, 0, 0, FLAGS, "sense"},
85 AVFILTER_DEFINE_CLASS(perspective);
87 static int query_formats(AVFilterContext *ctx)
89 static const enum AVPixelFormat pix_fmts[] = {
90 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA420P,
91 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ422P,AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ411P,
92 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
93 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP, AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE
96 ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
100 static inline double get_coeff(double d)
102 double coeff, A = -0.60;
107 coeff = (1.0 - (A + 3.0) * d * d + (A + 2.0) * d * d * d);
109 coeff = (-4.0 * A + 8.0 * A * d - 5.0 * A * d * d + A * d * d * d);
116 static const char *const var_names[] = { "W", "H", NULL };
117 enum { VAR_W, VAR_H, VAR_VARS_NB };
119 static int config_input(AVFilterLink *inlink)
121 double x0, x1, x2, x3, x4, x5, x6, x7, x8, q;
122 double t0, t1, t2, t3;
123 AVFilterContext *ctx = inlink->dst;
124 PerspectiveContext *s = ctx->priv;
125 double (*ref)[2] = s->ref;
126 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
127 double values[VAR_VARS_NB] = { [VAR_W] = inlink->w, [VAR_H] = inlink->h };
132 for (i = 0; i < 4; i++) {
133 for (j = 0; j < 2; j++) {
134 if (!s->expr_str[i][j])
135 return AVERROR(EINVAL);
136 ret = av_expr_parse_and_eval(&s->ref[i][j], s->expr_str[i][j],
137 var_names, &values[0],
138 NULL, NULL, NULL, NULL,
145 s->hsub = desc->log2_chroma_w;
146 s->vsub = desc->log2_chroma_h;
147 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
148 if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
151 s->height[1] = s->height[2] = FF_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
152 s->height[0] = s->height[3] = inlink->h;
154 s->pv = av_realloc_f(s->pv, w * h, 2 * sizeof(*s->pv));
156 return AVERROR(ENOMEM);
159 case PERSPECTIVE_SENSE_SOURCE:
160 x6 = ((ref[0][0] - ref[1][0] - ref[2][0] + ref[3][0]) *
161 (ref[2][1] - ref[3][1]) -
162 ( ref[0][1] - ref[1][1] - ref[2][1] + ref[3][1]) *
163 (ref[2][0] - ref[3][0])) * h;
164 x7 = ((ref[0][1] - ref[1][1] - ref[2][1] + ref[3][1]) *
165 (ref[1][0] - ref[3][0]) -
166 ( ref[0][0] - ref[1][0] - ref[2][0] + ref[3][0]) *
167 (ref[1][1] - ref[3][1])) * w;
168 q = ( ref[1][0] - ref[3][0]) * (ref[2][1] - ref[3][1]) -
169 ( ref[2][0] - ref[3][0]) * (ref[1][1] - ref[3][1]);
171 x0 = q * (ref[1][0] - ref[0][0]) * h + x6 * ref[1][0];
172 x1 = q * (ref[2][0] - ref[0][0]) * w + x7 * ref[2][0];
173 x2 = q * ref[0][0] * w * h;
174 x3 = q * (ref[1][1] - ref[0][1]) * h + x6 * ref[1][1];
175 x4 = q * (ref[2][1] - ref[0][1]) * w + x7 * ref[2][1];
176 x5 = q * ref[0][1] * w * h;
179 case PERSPECTIVE_SENSE_DESTINATION:
180 t0 = ref[0][0] * (ref[3][1] - ref[1][1]) +
181 ref[1][0] * (ref[0][1] - ref[3][1]) +
182 ref[3][0] * (ref[1][1] - ref[0][1]);
183 t1 = ref[1][0] * (ref[2][1] - ref[3][1]) +
184 ref[2][0] * (ref[3][1] - ref[1][1]) +
185 ref[3][0] * (ref[1][1] - ref[2][1]);
186 t2 = ref[0][0] * (ref[3][1] - ref[2][1]) +
187 ref[2][0] * (ref[0][1] - ref[3][1]) +
188 ref[3][0] * (ref[2][1] - ref[0][1]);
189 t3 = ref[0][0] * (ref[1][1] - ref[2][1]) +
190 ref[1][0] * (ref[2][1] - ref[0][1]) +
191 ref[2][0] * (ref[0][1] - ref[1][1]);
193 x0 = t0 * t1 * w * (ref[2][1] - ref[0][1]);
194 x1 = t0 * t1 * w * (ref[0][0] - ref[2][0]);
195 x2 = t0 * t1 * w * (ref[0][1] * ref[2][0] - ref[0][0] * ref[2][1]);
196 x3 = t1 * t2 * h * (ref[1][1] - ref[0][1]);
197 x4 = t1 * t2 * h * (ref[0][0] - ref[1][0]);
198 x5 = t1 * t2 * h * (ref[0][1] * ref[1][0] - ref[0][0] * ref[1][1]);
199 x6 = t1 * t2 * (ref[1][1] - ref[0][1]) +
200 t0 * t3 * (ref[2][1] - ref[3][1]);
201 x7 = t1 * t2 * (ref[0][0] - ref[1][0]) +
202 t0 * t3 * (ref[3][0] - ref[2][0]);
203 x8 = t1 * t2 * (ref[0][1] * ref[1][0] - ref[0][0] * ref[1][1]) +
204 t0 * t3 * (ref[2][0] * ref[3][1] - ref[2][1] * ref[3][0]);
210 for (y = 0; y < h; y++){
211 for (x = 0; x < w; x++){
214 u = (int)floor(SUB_PIXELS * (x0 * x + x1 * y + x2) /
215 (x6 * x + x7 * y + x8) + 0.5);
216 v = (int)floor(SUB_PIXELS * (x3 * x + x4 * y + x5) /
217 (x6 * x + x7 * y + x8) + 0.5);
219 s->pv[x + y * w][0] = u;
220 s->pv[x + y * w][1] = v;
224 for (i = 0; i < SUB_PIXELS; i++){
225 double d = i / (double)SUB_PIXELS;
229 for (j = 0; j < 4; j++)
230 temp[j] = get_coeff(j - d - 1);
232 for (j = 0; j < 4; j++)
235 for (j = 0; j < 4; j++)
236 s->coeff[i][j] = (int)floor((1 << COEFF_BITS) * temp[j] / sum + 0.5);
242 typedef struct ThreadData {
251 static int resample_cubic(AVFilterContext *ctx, void *arg,
252 int job, int nb_jobs)
254 PerspectiveContext *s = ctx->priv;
255 ThreadData *td = arg;
256 uint8_t *dst = td->dst;
257 int dst_linesize = td->dst_linesize;
258 uint8_t *src = td->src;
259 int src_linesize = td->src_linesize;
264 int start = (h * job) / nb_jobs;
265 int end = (h * (job+1)) / nb_jobs;
266 const int linesize = s->linesize[0];
269 for (y = start; y < end; y++) {
271 for (x = 0; x < w; x++) {
272 int u, v, subU, subV, sum, sx;
275 u = s->pv[sx + sy * linesize][0] >> hsub;
276 v = s->pv[sx + sy * linesize][1] >> vsub;
277 subU = u & (SUB_PIXELS - 1);
278 subV = v & (SUB_PIXELS - 1);
279 u >>= SUB_PIXEL_BITS;
280 v >>= SUB_PIXEL_BITS;
282 if (u > 0 && v > 0 && u < w - 2 && v < h - 2){
283 const int index = u + v*src_linesize;
284 const int a = s->coeff[subU][0];
285 const int b = s->coeff[subU][1];
286 const int c = s->coeff[subU][2];
287 const int d = s->coeff[subU][3];
289 sum = s->coeff[subV][0] * (a * src[index - 1 - src_linesize] + b * src[index - 0 - src_linesize] +
290 c * src[index + 1 - src_linesize] + d * src[index + 2 - src_linesize]) +
291 s->coeff[subV][1] * (a * src[index - 1 ] + b * src[index - 0 ] +
292 c * src[index + 1 ] + d * src[index + 2 ]) +
293 s->coeff[subV][2] * (a * src[index - 1 + src_linesize] + b * src[index - 0 + src_linesize] +
294 c * src[index + 1 + src_linesize] + d * src[index + 2 + src_linesize]) +
295 s->coeff[subV][3] * (a * src[index - 1 + 2 * src_linesize] + b * src[index - 0 + 2 * src_linesize] +
296 c * src[index + 1 + 2 * src_linesize] + d * src[index + 2 + 2 * src_linesize]);
302 for (dy = 0; dy < 4; dy++) {
309 for (dx = 0; dx < 4; dx++) {
317 sum += s->coeff[subU][dx] * s->coeff[subV][dy] * src[ ix + iy * src_linesize];
322 sum = (sum + (1<<(COEFF_BITS * 2 - 1))) >> (COEFF_BITS * 2);
323 sum = av_clip(sum, 0, 255);
324 dst[x + y * dst_linesize] = sum;
330 static int resample_linear(AVFilterContext *ctx, void *arg,
331 int job, int nb_jobs)
333 PerspectiveContext *s = ctx->priv;
334 ThreadData *td = arg;
335 uint8_t *dst = td->dst;
336 int dst_linesize = td->dst_linesize;
337 uint8_t *src = td->src;
338 int src_linesize = td->src_linesize;
343 int start = (h * job) / nb_jobs;
344 int end = (h * (job+1)) / nb_jobs;
345 const int linesize = s->linesize[0];
348 for (y = start; y < end; y++){
350 for (x = 0; x < w; x++){
351 int u, v, subU, subV, sum, sx, index, subUI, subVI;
354 u = s->pv[sx + sy * linesize][0] >> hsub;
355 v = s->pv[sx + sy * linesize][1] >> vsub;
356 subU = u & (SUB_PIXELS - 1);
357 subV = v & (SUB_PIXELS - 1);
358 u >>= SUB_PIXEL_BITS;
359 v >>= SUB_PIXEL_BITS;
361 index = u + v * src_linesize;
362 subUI = SUB_PIXELS - subU;
363 subVI = SUB_PIXELS - subV;
365 if ((unsigned)u < (unsigned)(w - 1)){
366 if((unsigned)v < (unsigned)(h - 1)){
367 sum = subVI * (subUI * src[index] + subU * src[index + 1]) +
368 subV * (subUI * src[index + src_linesize] + subU * src[index + src_linesize + 1]);
369 sum = (sum + (1 << (SUB_PIXEL_BITS * 2 - 1)))>> (SUB_PIXEL_BITS * 2);
375 index = u + v * src_linesize;
376 sum = subUI * src[index] + subU * src[index + 1];
377 sum = (sum + (1 << (SUB_PIXEL_BITS - 1))) >> SUB_PIXEL_BITS;
384 if ((unsigned)v < (unsigned)(h - 1)){
385 index = u + v * src_linesize;
386 sum = subVI * src[index] + subV * src[index + src_linesize];
387 sum = (sum + (1 << (SUB_PIXEL_BITS - 1))) >> SUB_PIXEL_BITS;
393 index = u + v * src_linesize;
398 sum = av_clip(sum, 0, 255);
399 dst[x + y * dst_linesize] = sum;
405 static av_cold int init(AVFilterContext *ctx)
407 PerspectiveContext *s = ctx->priv;
409 switch (s->interpolation) {
410 case LINEAR: s->perspective = resample_linear; break;
411 case CUBIC: s->perspective = resample_cubic; break;
417 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
419 AVFilterContext *ctx = inlink->dst;
420 AVFilterLink *outlink = ctx->outputs[0];
421 PerspectiveContext *s = ctx->priv;
425 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
427 av_frame_free(&frame);
428 return AVERROR(ENOMEM);
430 av_frame_copy_props(out, frame);
432 for (plane = 0; plane < s->nb_planes; plane++) {
433 int hsub = plane == 1 || plane == 2 ? s->hsub : 0;
434 int vsub = plane == 1 || plane == 2 ? s->vsub : 0;
435 ThreadData td = {.dst = out->data[plane],
436 .dst_linesize = out->linesize[plane],
437 .src = frame->data[plane],
438 .src_linesize = frame->linesize[plane],
439 .w = s->linesize[plane],
440 .h = s->height[plane],
443 ctx->internal->execute(ctx, s->perspective, &td, NULL, FFMIN(td.h, ctx->graph->nb_threads));
446 av_frame_free(&frame);
447 return ff_filter_frame(outlink, out);
450 static av_cold void uninit(AVFilterContext *ctx)
452 PerspectiveContext *s = ctx->priv;
457 static const AVFilterPad perspective_inputs[] = {
460 .type = AVMEDIA_TYPE_VIDEO,
461 .filter_frame = filter_frame,
462 .config_props = config_input,
467 static const AVFilterPad perspective_outputs[] = {
470 .type = AVMEDIA_TYPE_VIDEO,
475 AVFilter ff_vf_perspective = {
476 .name = "perspective",
477 .description = NULL_IF_CONFIG_SMALL("Correct the perspective of video."),
478 .priv_size = sizeof(PerspectiveContext),
481 .query_formats = query_formats,
482 .inputs = perspective_inputs,
483 .outputs = perspective_outputs,
484 .priv_class = &perspective_class,
485 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,