2 * Copyright (c) 2011 Smartjog S.A.S, Clément Bœsch <clement.boesch@smartjog.com>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * Potential thumbnail lookup filter to reduce the risk of an inappropriate
24 * selection (such as a black frame) we could get with an absolute seek.
26 * Simplified version of algorithm by Vadim Zaliva <lord@crocodile.org>.
27 * @see http://notbrainsurgery.livejournal.com/29773.html
33 #define HIST_SIZE (3*256)
36 AVFilterBufferRef *buf; ///< cached frame
37 int histogram[HIST_SIZE]; ///< RGB color distribution histogram of the frame
41 int n; ///< current frame
42 int n_frames; ///< number of frames for analysis
43 struct thumb_frame *frames; ///< the n_frames frames
46 static av_cold int init(AVFilterContext *ctx, const char *args)
48 ThumbContext *thumb = ctx->priv;
51 thumb->n_frames = 100;
53 int n = sscanf(args, "%d", &thumb->n_frames);
54 if (n != 1 || thumb->n_frames < 2) {
56 av_log(ctx, AV_LOG_ERROR,
57 "Invalid number of frames specified (minimum is 2).\n");
58 return AVERROR(EINVAL);
61 thumb->frames = av_calloc(thumb->n_frames, sizeof(*thumb->frames));
63 av_log(ctx, AV_LOG_ERROR,
64 "Allocation failure, try to lower the number of frames\n");
65 return AVERROR(ENOMEM);
67 av_log(ctx, AV_LOG_VERBOSE, "batch size: %d frames\n", thumb->n_frames);
72 * @brief Compute Sum-square deviation to estimate "closeness".
73 * @param hist color distribution histogram
74 * @param median average color distribution histogram
75 * @return sum of squared errors
77 static double frame_sum_square_err(const int *hist, const double *median)
80 double err, sum_sq_err = 0;
82 for (i = 0; i < HIST_SIZE; i++) {
83 err = median[i] - (double)hist[i];
84 sum_sq_err += err*err;
89 static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *frame)
91 int i, j, best_frame_idx = 0;
92 double avg_hist[HIST_SIZE] = {0}, sq_err, min_sq_err = -1;
93 AVFilterContext *ctx = inlink->dst;
94 ThumbContext *thumb = ctx->priv;
95 AVFilterLink *outlink = ctx->outputs[0];
96 AVFilterBufferRef *picref;
97 int *hist = thumb->frames[thumb->n].histogram;
98 const uint8_t *p = frame->data[0];
100 // keep a reference of each frame
101 thumb->frames[thumb->n].buf = frame;
103 // update current frame RGB histogram
104 for (j = 0; j < inlink->h; j++) {
105 for (i = 0; i < inlink->w; i++) {
106 hist[0*256 + p[i*3 ]]++;
107 hist[1*256 + p[i*3 + 1]]++;
108 hist[2*256 + p[i*3 + 2]]++;
110 p += frame->linesize[0];
113 // no selection until the buffer of N frames is filled up
114 if (thumb->n < thumb->n_frames - 1) {
119 // average histogram of the N frames
120 for (j = 0; j < FF_ARRAY_ELEMS(avg_hist); j++) {
121 for (i = 0; i < thumb->n_frames; i++)
122 avg_hist[j] += (double)thumb->frames[i].histogram[j];
123 avg_hist[j] /= thumb->n_frames;
126 // find the frame closer to the average using the sum of squared errors
127 for (i = 0; i < thumb->n_frames; i++) {
128 sq_err = frame_sum_square_err(thumb->frames[i].histogram, avg_hist);
129 if (i == 0 || sq_err < min_sq_err)
130 best_frame_idx = i, min_sq_err = sq_err;
133 // free and reset everything (except the best frame buffer)
134 for (i = 0; i < thumb->n_frames; i++) {
135 memset(thumb->frames[i].histogram, 0, sizeof(thumb->frames[i].histogram));
136 if (i == best_frame_idx)
138 avfilter_unref_bufferp(&thumb->frames[i].buf);
142 // raise the chosen one
143 picref = thumb->frames[best_frame_idx].buf;
144 av_log(ctx, AV_LOG_INFO, "frame id #%d (pts_time=%f) selected\n",
145 best_frame_idx, picref->pts * av_q2d(inlink->time_base));
146 thumb->frames[best_frame_idx].buf = NULL;
147 return ff_filter_frame(outlink, picref);
150 static av_cold void uninit(AVFilterContext *ctx)
153 ThumbContext *thumb = ctx->priv;
154 for (i = 0; i < thumb->n_frames && thumb->frames[i].buf; i++)
155 avfilter_unref_bufferp(&thumb->frames[i].buf);
156 av_freep(&thumb->frames);
159 static int request_frame(AVFilterLink *link)
161 ThumbContext *thumb = link->src->priv;
163 /* loop until a frame thumbnail is available (when a frame is queued,
164 * thumb->n is reset to zero) */
166 int ret = ff_request_frame(link->src->inputs[0]);
173 static int poll_frame(AVFilterLink *link)
175 ThumbContext *thumb = link->src->priv;
176 AVFilterLink *inlink = link->src->inputs[0];
177 int ret, available_frames = ff_poll_frame(inlink);
179 /* If the input link is not able to provide any frame, we can't do anything
180 * at the moment and thus have zero thumbnail available. */
181 if (!available_frames)
184 /* Since at least one frame is available and the next frame will allow us
185 * to compute a thumbnail, we can return 1 frame. */
186 if (thumb->n == thumb->n_frames - 1)
189 /* we have some frame(s) available in the input link, but not yet enough to
190 * output a thumbnail, so we request more */
191 ret = ff_request_frame(inlink);
192 return ret < 0 ? ret : 0;
195 static int query_formats(AVFilterContext *ctx)
197 static const enum AVPixelFormat pix_fmts[] = {
198 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
201 ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
205 static const AVFilterPad thumbnail_inputs[] = {
208 .type = AVMEDIA_TYPE_VIDEO,
209 .get_video_buffer = ff_null_get_video_buffer,
210 .min_perms = AV_PERM_PRESERVE,
211 .filter_frame = filter_frame,
216 static const AVFilterPad thumbnail_outputs[] = {
219 .type = AVMEDIA_TYPE_VIDEO,
220 .request_frame = request_frame,
221 .poll_frame = poll_frame,
226 AVFilter avfilter_vf_thumbnail = {
228 .description = NULL_IF_CONFIG_SMALL("Select the most representative frame in a given sequence of consecutive frames."),
229 .priv_size = sizeof(ThumbContext),
232 .query_formats = query_formats,
233 .inputs = thumbnail_inputs,
234 .outputs = thumbnail_outputs,