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/eval.h"
23 #include "libavutil/imgutils.h"
24 #include "libavutil/pixdesc.h"
25 #include "libavutil/opt.h"
31 #define SUB_PIXEL_BITS 8
32 #define SUB_PIXELS (1 << SUB_PIXEL_BITS)
38 typedef struct PerspectiveContext {
43 int32_t coeff[SUB_PIXELS][4];
50 void (*perspective)(struct PerspectiveContext *s,
51 uint8_t *dst, int dst_linesize,
52 uint8_t *src, int src_linesize,
53 int w, int h, int hsub, int vsub);
56 #define OFFSET(x) offsetof(PerspectiveContext, x)
57 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
59 static const AVOption perspective_options[] = {
60 { "x0", "set top left x coordinate", OFFSET(expr_str[0][0]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
61 { "y0", "set top left y coordinate", OFFSET(expr_str[0][1]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
62 { "x1", "set top right x coordinate", OFFSET(expr_str[1][0]), AV_OPT_TYPE_STRING, {.str="W"}, 0, 0, FLAGS },
63 { "y1", "set top right y coordinate", OFFSET(expr_str[1][1]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
64 { "x2", "set bottom left x coordinate", OFFSET(expr_str[2][0]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
65 { "y2", "set bottom left y coordinate", OFFSET(expr_str[2][1]), AV_OPT_TYPE_STRING, {.str="H"}, 0, 0, FLAGS },
66 { "x3", "set bottom right x coordinate", OFFSET(expr_str[3][0]), AV_OPT_TYPE_STRING, {.str="W"}, 0, 0, FLAGS },
67 { "y3", "set bottom right y coordinate", OFFSET(expr_str[3][1]), AV_OPT_TYPE_STRING, {.str="H"}, 0, 0, FLAGS },
68 { "interpolation", "set interpolation", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=LINEAR}, 0, 1, FLAGS, "interpolation" },
69 { "linear", "", 0, AV_OPT_TYPE_CONST, {.i64=LINEAR}, 0, 0, FLAGS, "interpolation" },
70 { "cubic", "", 0, AV_OPT_TYPE_CONST, {.i64=CUBIC}, 0, 0, FLAGS, "interpolation" },
74 AVFILTER_DEFINE_CLASS(perspective);
76 static int query_formats(AVFilterContext *ctx)
78 static const enum AVPixelFormat pix_fmts[] = {
79 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA420P,
80 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ422P,AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ411P,
81 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
82 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP, AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE
85 ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
89 static inline double get_coeff(double d)
91 double coeff, A = -0.60;
96 coeff = (1.0 - (A + 3.0) * d * d + (A + 2.0) * d * d * d);
98 coeff = (-4.0 * A + 8.0 * A * d - 5.0 * A * d * d + A * d * d * d);
105 static const char *const var_names[] = { "W", "H", NULL };
106 enum { VAR_W, VAR_H, VAR_VARS_NB };
108 static int config_input(AVFilterLink *inlink)
110 double x0, x1, x2, x3, x4, x5, x6, x7, q;
111 AVFilterContext *ctx = inlink->dst;
112 PerspectiveContext *s = ctx->priv;
113 double (*ref)[2] = s->ref;
114 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
115 double values[VAR_VARS_NB] = { [VAR_W] = inlink->w, [VAR_H] = inlink->h };
120 for (i = 0; i < 4; i++) {
121 for (j = 0; j < 2; j++) {
122 if (!s->expr_str[i][j])
123 return AVERROR(EINVAL);
124 ret = av_expr_parse_and_eval(&s->ref[i][j], s->expr_str[i][j],
125 var_names, &values[0],
126 NULL, NULL, NULL, NULL,
133 s->hsub = desc->log2_chroma_w;
134 s->vsub = desc->log2_chroma_h;
135 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
136 if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
139 s->height[1] = s->height[2] = FF_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
140 s->height[0] = s->height[3] = inlink->h;
142 s->pv = av_realloc_f(s->pv, w * h, 2 * sizeof(*s->pv));
144 return AVERROR(ENOMEM);
146 x6 = ((ref[0][0] - ref[1][0] - ref[2][0] + ref[3][0]) *
147 (ref[2][1] - ref[3][1]) -
148 ( ref[0][1] - ref[1][1] - ref[2][1] + ref[3][1]) *
149 (ref[2][0] - ref[3][0])) * h;
150 x7 = ((ref[0][1] - ref[1][1] - ref[2][1] + ref[3][1]) *
151 (ref[1][0] - ref[3][0]) -
152 ( ref[0][0] - ref[1][0] - ref[2][0] + ref[3][0]) *
153 (ref[1][1] - ref[3][1])) * w;
154 q = ( ref[1][0] - ref[3][0]) * (ref[2][1] - ref[3][1]) -
155 ( ref[2][0] - ref[3][0]) * (ref[1][1] - ref[3][1]);
157 x0 = q * (ref[1][0] - ref[0][0]) * h + x6 * ref[1][0];
158 x1 = q * (ref[2][0] - ref[0][0]) * w + x7 * ref[2][0];
159 x2 = q * ref[0][0] * w * h;
160 x3 = q * (ref[1][1] - ref[0][1]) * h + x6 * ref[1][1];
161 x4 = q * (ref[2][1] - ref[0][1]) * w + x7 * ref[2][1];
162 x5 = q * ref[0][1] * w * h;
164 for (y = 0; y < h; y++){
165 for (x = 0; x < w; x++){
168 u = (int)floor(SUB_PIXELS * (x0 * x + x1 * y + x2) /
169 (x6 * x + x7 * y + q * w * h) + 0.5);
170 v = (int)floor(SUB_PIXELS * (x3 * x + x4 * y + x5) /
171 (x6 * x + x7 * y + q * w * h) + 0.5);
173 s->pv[x + y * w][0] = u;
174 s->pv[x + y * w][1] = v;
178 for (i = 0; i < SUB_PIXELS; i++){
179 double d = i / (double)SUB_PIXELS;
183 for (j = 0; j < 4; j++)
184 temp[j] = get_coeff(j - d - 1);
186 for (j = 0; j < 4; j++)
189 for (j = 0; j < 4; j++)
190 s->coeff[i][j] = (int)floor((1 << COEFF_BITS) * temp[j] / sum + 0.5);
196 static void resample_cubic(PerspectiveContext *s,
197 uint8_t *dst, int dst_linesize,
198 uint8_t *src, int src_linesize,
199 int w, int h, int hsub, int vsub)
201 const int linesize = s->linesize[0];
204 for (y = 0; y < h; y++) {
206 for (x = 0; x < w; x++) {
207 int u, v, subU, subV, sum, sx;
210 u = s->pv[sx + sy * linesize][0] >> hsub;
211 v = s->pv[sx + sy * linesize][1] >> vsub;
212 subU = u & (SUB_PIXELS - 1);
213 subV = v & (SUB_PIXELS - 1);
214 u >>= SUB_PIXEL_BITS;
215 v >>= SUB_PIXEL_BITS;
217 if (u > 0 && v > 0 && u < w - 2 && v < h - 2){
218 const int index = u + v*src_linesize;
219 const int a = s->coeff[subU][0];
220 const int b = s->coeff[subU][1];
221 const int c = s->coeff[subU][2];
222 const int d = s->coeff[subU][3];
224 sum = s->coeff[subV][0] * (a * src[index - 1 - src_linesize] + b * src[index - 0 - src_linesize] +
225 c * src[index + 1 - src_linesize] + d * src[index + 2 - src_linesize]) +
226 s->coeff[subV][1] * (a * src[index - 1 ] + b * src[index - 0 ] +
227 c * src[index + 1 ] + d * src[index + 2 ]) +
228 s->coeff[subV][2] * (a * src[index - 1 + src_linesize] + b * src[index - 0 + src_linesize] +
229 c * src[index + 1 + src_linesize] + d * src[index + 2 + src_linesize]) +
230 s->coeff[subV][3] * (a * src[index - 1 + 2 * src_linesize] + b * src[index - 0 + 2 * src_linesize] +
231 c * src[index + 1 + 2 * src_linesize] + d * src[index + 2 + 2 * src_linesize]);
237 for (dy = 0; dy < 4; dy++) {
244 for (dx = 0; dx < 4; dx++) {
252 sum += s->coeff[subU][dx] * s->coeff[subV][dy] * src[ ix + iy * src_linesize];
257 sum = (sum + (1<<(COEFF_BITS * 2 - 1))) >> (COEFF_BITS * 2);
258 sum = av_clip(sum, 0, 255);
259 dst[x + y * dst_linesize] = sum;
264 static void resample_linear(PerspectiveContext *s,
265 uint8_t *dst, int dst_linesize,
266 uint8_t *src, int src_linesize,
267 int w, int h, int hsub, int vsub)
269 const int linesize = s->linesize[0];
272 for (y = 0; y < h; y++){
274 for (x = 0; x < w; x++){
275 int u, v, subU, subV, sum, sx, index, subUI, subVI;
278 u = s->pv[sx + sy * linesize][0] >> hsub;
279 v = s->pv[sx + sy * linesize][1] >> vsub;
280 subU = u & (SUB_PIXELS - 1);
281 subV = v & (SUB_PIXELS - 1);
282 u >>= SUB_PIXEL_BITS;
283 v >>= SUB_PIXEL_BITS;
285 index = u + v * src_linesize;
286 subUI = SUB_PIXELS - subU;
287 subVI = SUB_PIXELS - subV;
289 if ((unsigned)u < (unsigned)(w - 1)){
290 if((unsigned)v < (unsigned)(h - 1)){
291 sum = subVI * (subUI * src[index] + subU * src[index + 1]) +
292 subV * (subUI * src[index + src_linesize] + subU * src[index + src_linesize + 1]);
293 sum = (sum + (1 << (SUB_PIXEL_BITS * 2 - 1)))>> (SUB_PIXEL_BITS * 2);
299 index = u + v * src_linesize;
300 sum = subUI * src[index] + subU * src[index + 1];
301 sum = (sum + (1 << (SUB_PIXEL_BITS - 1))) >> SUB_PIXEL_BITS;
308 if ((unsigned)v < (unsigned)(h - 1)){
309 index = u + v * src_linesize;
310 sum = subVI * src[index] + subV * src[index + src_linesize];
311 sum = (sum + (1 << (SUB_PIXEL_BITS - 1))) >> SUB_PIXEL_BITS;
317 index = u + v * src_linesize;
322 sum = av_clip(sum, 0, 255);
323 dst[x + y * dst_linesize] = sum;
328 static av_cold int init(AVFilterContext *ctx)
330 PerspectiveContext *s = ctx->priv;
332 switch (s->interpolation) {
333 case LINEAR: s->perspective = resample_linear; break;
334 case CUBIC: s->perspective = resample_cubic; break;
340 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
342 AVFilterContext *ctx = inlink->dst;
343 AVFilterLink *outlink = ctx->outputs[0];
344 PerspectiveContext *s = ctx->priv;
348 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
350 av_frame_free(&frame);
351 return AVERROR(ENOMEM);
353 av_frame_copy_props(out, frame);
355 for (plane = 0; plane < s->nb_planes; plane++) {
356 int hsub = plane == 1 || plane == 2 ? s->hsub : 0;
357 int vsub = plane == 1 || plane == 2 ? s->vsub : 0;
358 s->perspective(s, out->data[plane], out->linesize[plane],
359 frame->data[plane], frame->linesize[plane],
360 s->linesize[plane], s->height[plane], hsub, vsub);
363 av_frame_free(&frame);
364 return ff_filter_frame(outlink, out);
367 static av_cold void uninit(AVFilterContext *ctx)
369 PerspectiveContext *s = ctx->priv;
374 static const AVFilterPad perspective_inputs[] = {
377 .type = AVMEDIA_TYPE_VIDEO,
378 .filter_frame = filter_frame,
379 .config_props = config_input,
384 static const AVFilterPad perspective_outputs[] = {
387 .type = AVMEDIA_TYPE_VIDEO,
392 AVFilter ff_vf_perspective = {
393 .name = "perspective",
394 .description = NULL_IF_CONFIG_SMALL("Correct the perspective of video."),
395 .priv_size = sizeof(PerspectiveContext),
398 .query_formats = query_formats,
399 .inputs = perspective_inputs,
400 .outputs = perspective_outputs,
401 .priv_class = &perspective_class,
402 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,