| 1 | /* |
| 2 | * Copyright (c) 2011 Smartjog S.A.S, Clément Bœsch <clement.boesch@smartjog.com> |
| 3 | * |
| 4 | * This file is part of FFmpeg. |
| 5 | * |
| 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. |
| 10 | * |
| 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. |
| 15 | * |
| 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 |
| 19 | */ |
| 20 | |
| 21 | /** |
| 22 | * @file |
| 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. |
| 25 | * |
| 26 | * Simplified version of algorithm by Vadim Zaliva <lord@crocodile.org>. |
| 27 | * @see http://notbrainsurgery.livejournal.com/29773.html |
| 28 | */ |
| 29 | |
| 30 | #include "libavutil/opt.h" |
| 31 | #include "avfilter.h" |
| 32 | #include "internal.h" |
| 33 | |
| 34 | #define HIST_SIZE (3*256) |
| 35 | |
| 36 | struct thumb_frame { |
| 37 | AVFrame *buf; ///< cached frame |
| 38 | int histogram[HIST_SIZE]; ///< RGB color distribution histogram of the frame |
| 39 | }; |
| 40 | |
| 41 | typedef struct { |
| 42 | const AVClass *class; |
| 43 | int n; ///< current frame |
| 44 | int n_frames; ///< number of frames for analysis |
| 45 | struct thumb_frame *frames; ///< the n_frames frames |
| 46 | AVRational tb; ///< copy of the input timebase to ease access |
| 47 | } ThumbContext; |
| 48 | |
| 49 | #define OFFSET(x) offsetof(ThumbContext, x) |
| 50 | #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM |
| 51 | |
| 52 | static const AVOption thumbnail_options[] = { |
| 53 | { "n", "set the frames batch size", OFFSET(n_frames), AV_OPT_TYPE_INT, {.i64=100}, 2, INT_MAX, FLAGS }, |
| 54 | { NULL } |
| 55 | }; |
| 56 | |
| 57 | AVFILTER_DEFINE_CLASS(thumbnail); |
| 58 | |
| 59 | static av_cold int init(AVFilterContext *ctx) |
| 60 | { |
| 61 | ThumbContext *thumb = ctx->priv; |
| 62 | |
| 63 | thumb->frames = av_calloc(thumb->n_frames, sizeof(*thumb->frames)); |
| 64 | if (!thumb->frames) { |
| 65 | av_log(ctx, AV_LOG_ERROR, |
| 66 | "Allocation failure, try to lower the number of frames\n"); |
| 67 | return AVERROR(ENOMEM); |
| 68 | } |
| 69 | av_log(ctx, AV_LOG_VERBOSE, "batch size: %d frames\n", thumb->n_frames); |
| 70 | return 0; |
| 71 | } |
| 72 | |
| 73 | /** |
| 74 | * @brief Compute Sum-square deviation to estimate "closeness". |
| 75 | * @param hist color distribution histogram |
| 76 | * @param median average color distribution histogram |
| 77 | * @return sum of squared errors |
| 78 | */ |
| 79 | static double frame_sum_square_err(const int *hist, const double *median) |
| 80 | { |
| 81 | int i; |
| 82 | double err, sum_sq_err = 0; |
| 83 | |
| 84 | for (i = 0; i < HIST_SIZE; i++) { |
| 85 | err = median[i] - (double)hist[i]; |
| 86 | sum_sq_err += err*err; |
| 87 | } |
| 88 | return sum_sq_err; |
| 89 | } |
| 90 | |
| 91 | static AVFrame *get_best_frame(AVFilterContext *ctx) |
| 92 | { |
| 93 | AVFrame *picref; |
| 94 | ThumbContext *thumb = ctx->priv; |
| 95 | int i, j, best_frame_idx = 0; |
| 96 | int nb_frames = thumb->n; |
| 97 | double avg_hist[HIST_SIZE] = {0}, sq_err, min_sq_err = -1; |
| 98 | |
| 99 | // average histogram of the N frames |
| 100 | for (j = 0; j < FF_ARRAY_ELEMS(avg_hist); j++) { |
| 101 | for (i = 0; i < nb_frames; i++) |
| 102 | avg_hist[j] += (double)thumb->frames[i].histogram[j]; |
| 103 | avg_hist[j] /= nb_frames; |
| 104 | } |
| 105 | |
| 106 | // find the frame closer to the average using the sum of squared errors |
| 107 | for (i = 0; i < nb_frames; i++) { |
| 108 | sq_err = frame_sum_square_err(thumb->frames[i].histogram, avg_hist); |
| 109 | if (i == 0 || sq_err < min_sq_err) |
| 110 | best_frame_idx = i, min_sq_err = sq_err; |
| 111 | } |
| 112 | |
| 113 | // free and reset everything (except the best frame buffer) |
| 114 | for (i = 0; i < nb_frames; i++) { |
| 115 | memset(thumb->frames[i].histogram, 0, sizeof(thumb->frames[i].histogram)); |
| 116 | if (i != best_frame_idx) |
| 117 | av_frame_free(&thumb->frames[i].buf); |
| 118 | } |
| 119 | thumb->n = 0; |
| 120 | |
| 121 | // raise the chosen one |
| 122 | picref = thumb->frames[best_frame_idx].buf; |
| 123 | av_log(ctx, AV_LOG_INFO, "frame id #%d (pts_time=%f) selected " |
| 124 | "from a set of %d images\n", best_frame_idx, |
| 125 | picref->pts * av_q2d(thumb->tb), nb_frames); |
| 126 | thumb->frames[best_frame_idx].buf = NULL; |
| 127 | |
| 128 | return picref; |
| 129 | } |
| 130 | |
| 131 | static int filter_frame(AVFilterLink *inlink, AVFrame *frame) |
| 132 | { |
| 133 | int i, j; |
| 134 | AVFilterContext *ctx = inlink->dst; |
| 135 | ThumbContext *thumb = ctx->priv; |
| 136 | AVFilterLink *outlink = ctx->outputs[0]; |
| 137 | int *hist = thumb->frames[thumb->n].histogram; |
| 138 | const uint8_t *p = frame->data[0]; |
| 139 | |
| 140 | // keep a reference of each frame |
| 141 | thumb->frames[thumb->n].buf = frame; |
| 142 | |
| 143 | // update current frame RGB histogram |
| 144 | for (j = 0; j < inlink->h; j++) { |
| 145 | for (i = 0; i < inlink->w; i++) { |
| 146 | hist[0*256 + p[i*3 ]]++; |
| 147 | hist[1*256 + p[i*3 + 1]]++; |
| 148 | hist[2*256 + p[i*3 + 2]]++; |
| 149 | } |
| 150 | p += frame->linesize[0]; |
| 151 | } |
| 152 | |
| 153 | // no selection until the buffer of N frames is filled up |
| 154 | thumb->n++; |
| 155 | if (thumb->n < thumb->n_frames) |
| 156 | return 0; |
| 157 | |
| 158 | return ff_filter_frame(outlink, get_best_frame(ctx)); |
| 159 | } |
| 160 | |
| 161 | static av_cold void uninit(AVFilterContext *ctx) |
| 162 | { |
| 163 | int i; |
| 164 | ThumbContext *thumb = ctx->priv; |
| 165 | for (i = 0; i < thumb->n_frames && thumb->frames[i].buf; i++) |
| 166 | av_frame_free(&thumb->frames[i].buf); |
| 167 | av_freep(&thumb->frames); |
| 168 | } |
| 169 | |
| 170 | static int request_frame(AVFilterLink *link) |
| 171 | { |
| 172 | AVFilterContext *ctx = link->src; |
| 173 | ThumbContext *thumb = ctx->priv; |
| 174 | |
| 175 | /* loop until a frame thumbnail is available (when a frame is queued, |
| 176 | * thumb->n is reset to zero) */ |
| 177 | do { |
| 178 | int ret = ff_request_frame(ctx->inputs[0]); |
| 179 | if (ret == AVERROR_EOF && thumb->n) { |
| 180 | ret = ff_filter_frame(link, get_best_frame(ctx)); |
| 181 | if (ret < 0) |
| 182 | return ret; |
| 183 | ret = AVERROR_EOF; |
| 184 | } |
| 185 | if (ret < 0) |
| 186 | return ret; |
| 187 | } while (thumb->n); |
| 188 | return 0; |
| 189 | } |
| 190 | |
| 191 | static int config_props(AVFilterLink *inlink) |
| 192 | { |
| 193 | AVFilterContext *ctx = inlink->dst; |
| 194 | ThumbContext *thumb = ctx->priv; |
| 195 | |
| 196 | thumb->tb = inlink->time_base; |
| 197 | return 0; |
| 198 | } |
| 199 | |
| 200 | static int query_formats(AVFilterContext *ctx) |
| 201 | { |
| 202 | static const enum AVPixelFormat pix_fmts[] = { |
| 203 | AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24, |
| 204 | AV_PIX_FMT_NONE |
| 205 | }; |
| 206 | ff_set_common_formats(ctx, ff_make_format_list(pix_fmts)); |
| 207 | return 0; |
| 208 | } |
| 209 | |
| 210 | static const AVFilterPad thumbnail_inputs[] = { |
| 211 | { |
| 212 | .name = "default", |
| 213 | .type = AVMEDIA_TYPE_VIDEO, |
| 214 | .config_props = config_props, |
| 215 | .filter_frame = filter_frame, |
| 216 | }, |
| 217 | { NULL } |
| 218 | }; |
| 219 | |
| 220 | static const AVFilterPad thumbnail_outputs[] = { |
| 221 | { |
| 222 | .name = "default", |
| 223 | .type = AVMEDIA_TYPE_VIDEO, |
| 224 | .request_frame = request_frame, |
| 225 | }, |
| 226 | { NULL } |
| 227 | }; |
| 228 | |
| 229 | AVFilter ff_vf_thumbnail = { |
| 230 | .name = "thumbnail", |
| 231 | .description = NULL_IF_CONFIG_SMALL("Select the most representative frame in a given sequence of consecutive frames."), |
| 232 | .priv_size = sizeof(ThumbContext), |
| 233 | .init = init, |
| 234 | .uninit = uninit, |
| 235 | .query_formats = query_formats, |
| 236 | .inputs = thumbnail_inputs, |
| 237 | .outputs = thumbnail_outputs, |
| 238 | .priv_class = &thumbnail_class, |
| 239 | }; |