2 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (c) 2013 Clément Bœsch <u pkh me>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (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 Foundation, Inc.,
19 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * Simple post processing filter
26 * This implementation is based on an algorithm described in
27 * "Aria Nosratinia Embedded Post-Processing for
28 * Enhancement of Compressed Images (1999)"
30 * Originally written by Michael Niedermayer for the MPlayer project, and
31 * ported by Clément Bœsch for FFmpeg.
34 #include "libavutil/avassert.h"
35 #include "libavutil/imgutils.h"
36 #include "libavutil/mem_internal.h"
37 #include "libavutil/opt.h"
38 #include "libavutil/pixdesc.h"
49 static const AVClass *child_class_iterate(void **iter)
51 const AVClass *c = *iter ? NULL : avcodec_dct_get_class();
52 *iter = (void*)(uintptr_t)c;
56 static void *child_next(void *obj, void *prev)
59 return prev ? NULL : s->dct;
62 #define OFFSET(x) offsetof(SPPContext, x)
63 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
64 #define TFLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
65 static const AVOption spp_options[] = {
66 { "quality", "set quality", OFFSET(log2_count), AV_OPT_TYPE_INT, {.i64 = 3}, 0, MAX_LEVEL, TFLAGS },
67 { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 63, FLAGS },
68 { "mode", "set thresholding mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64 = MODE_HARD}, 0, NB_MODES - 1, FLAGS, "mode" },
69 { "hard", "hard thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_HARD}, INT_MIN, INT_MAX, FLAGS, "mode" },
70 { "soft", "soft thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_SOFT}, INT_MIN, INT_MAX, FLAGS, "mode" },
71 { "use_bframe_qp", "use B-frames' QP", OFFSET(use_bframe_qp), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, FLAGS },
75 static const AVClass spp_class = {
77 .item_name = av_default_item_name,
78 .option = spp_options,
79 .version = LIBAVUTIL_VERSION_INT,
80 .category = AV_CLASS_CATEGORY_FILTER,
81 .child_class_iterate = child_class_iterate,
82 .child_next = child_next,
85 // XXX: share between filters?
86 DECLARE_ALIGNED(8, static const uint8_t, ldither)[8][8] = {
87 { 0, 48, 12, 60, 3, 51, 15, 63 },
88 { 32, 16, 44, 28, 35, 19, 47, 31 },
89 { 8, 56, 4, 52, 11, 59, 7, 55 },
90 { 40, 24, 36, 20, 43, 27, 39, 23 },
91 { 2, 50, 14, 62, 1, 49, 13, 61 },
92 { 34, 18, 46, 30, 33, 17, 45, 29 },
93 { 10, 58, 6, 54, 9, 57, 5, 53 },
94 { 42, 26, 38, 22, 41, 25, 37, 21 },
97 static const uint8_t offset[127][2] = {
99 {0,0}, {4,4}, // quality = 1
100 {0,0}, {2,2}, {6,4}, {4,6}, // quality = 2
101 {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7}, // quality = 3
103 {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3}, // quality = 4
104 {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7},
106 {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7}, // quality = 5
107 {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7},
108 {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7},
109 {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7},
111 {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2}, // quality = 6
112 {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0},
113 {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3},
114 {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1},
115 {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3},
116 {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1},
117 {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2},
118 {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0},
121 static void hardthresh_c(int16_t dst[64], const int16_t src[64],
122 int qp, const uint8_t *permutation)
125 int bias = 0; // FIXME
127 unsigned threshold1 = qp * ((1<<4) - bias) - 1;
128 unsigned threshold2 = threshold1 << 1;
130 memset(dst, 0, 64 * sizeof(dst[0]));
131 dst[0] = (src[0] + 4) >> 3;
133 for (i = 1; i < 64; i++) {
135 if (((unsigned)(level + threshold1)) > threshold2) {
136 const int j = permutation[i];
137 dst[j] = (level + 4) >> 3;
142 static void softthresh_c(int16_t dst[64], const int16_t src[64],
143 int qp, const uint8_t *permutation)
146 int bias = 0; //FIXME
148 unsigned threshold1 = qp * ((1<<4) - bias) - 1;
149 unsigned threshold2 = threshold1 << 1;
151 memset(dst, 0, 64 * sizeof(dst[0]));
152 dst[0] = (src[0] + 4) >> 3;
154 for (i = 1; i < 64; i++) {
156 if (((unsigned)(level + threshold1)) > threshold2) {
157 const int j = permutation[i];
158 if (level > 0) dst[j] = (level - threshold1 + 4) >> 3;
159 else dst[j] = (level + threshold1 + 4) >> 3;
164 static void store_slice_c(uint8_t *dst, const int16_t *src,
165 int dst_linesize, int src_linesize,
166 int width, int height, int log2_scale,
167 const uint8_t dither[8][8])
171 #define STORE(pos) do { \
172 temp = ((src[x + y*src_linesize + pos] << log2_scale) + d[pos]) >> 6; \
174 temp = ~(temp >> 31); \
175 dst[x + y*dst_linesize + pos] = temp; \
178 for (y = 0; y < height; y++) {
179 const uint8_t *d = dither[y];
180 for (x = 0; x < width; x += 8) {
194 static void store_slice16_c(uint16_t *dst, const int16_t *src,
195 int dst_linesize, int src_linesize,
196 int width, int height, int log2_scale,
197 const uint8_t dither[8][8], int depth)
200 unsigned int mask = -1<<depth;
202 #define STORE16(pos) do { \
203 temp = ((src[x + y*src_linesize + pos] << log2_scale) + (d[pos]>>1)) >> 5; \
205 temp = ~(temp >> 31); \
206 dst[x + y*dst_linesize + pos] = temp; \
209 for (y = 0; y < height; y++) {
210 const uint8_t *d = dither[y];
211 for (x = 0; x < width; x += 8) {
225 static inline void add_block(uint16_t *dst, int linesize, const int16_t block[64])
229 for (y = 0; y < 8; y++) {
230 dst[0 + y*linesize] += block[0 + y*8];
231 dst[1 + y*linesize] += block[1 + y*8];
232 dst[2 + y*linesize] += block[2 + y*8];
233 dst[3 + y*linesize] += block[3 + y*8];
234 dst[4 + y*linesize] += block[4 + y*8];
235 dst[5 + y*linesize] += block[5 + y*8];
236 dst[6 + y*linesize] += block[6 + y*8];
237 dst[7 + y*linesize] += block[7 + y*8];
241 static void filter(SPPContext *p, uint8_t *dst, uint8_t *src,
242 int dst_linesize, int src_linesize, int width, int height,
243 const uint8_t *qp_table, int qp_stride, int is_luma, int depth)
246 const int count = 1 << p->log2_count;
247 const int linesize = is_luma ? p->temp_linesize : FFALIGN(width+16, 16);
248 DECLARE_ALIGNED(16, uint64_t, block_align)[32];
249 int16_t *block = (int16_t *)block_align;
250 int16_t *block2 = (int16_t *)(block_align + 16);
251 uint16_t *psrc16 = (uint16_t*)p->src;
252 const int sample_bytes = (depth+7) / 8;
254 for (y = 0; y < height; y++) {
255 int index = 8 + 8*linesize + y*linesize;
256 memcpy(p->src + index*sample_bytes, src + y*src_linesize, width*sample_bytes);
257 if (sample_bytes == 1) {
258 for (x = 0; x < 8; x++) {
259 p->src[index - x - 1] = p->src[index + x ];
260 p->src[index + width + x ] = p->src[index + width - x - 1];
263 for (x = 0; x < 8; x++) {
264 psrc16[index - x - 1] = psrc16[index + x ];
265 psrc16[index + width + x ] = psrc16[index + width - x - 1];
269 for (y = 0; y < 8; y++) {
270 memcpy(p->src + ( 7-y)*linesize * sample_bytes, p->src + ( y+8)*linesize * sample_bytes, linesize * sample_bytes);
271 memcpy(p->src + (height+8+y)*linesize * sample_bytes, p->src + (height-y+7)*linesize * sample_bytes, linesize * sample_bytes);
274 for (y = 0; y < height + 8; y += 8) {
275 memset(p->temp + (8 + y) * linesize, 0, 8 * linesize * sizeof(*p->temp));
276 for (x = 0; x < width + 8; x += 8) {
282 const int qps = 3 + is_luma;
283 qp = qp_table[(FFMIN(x, width - 1) >> qps) + (FFMIN(y, height - 1) >> qps) * qp_stride];
284 qp = FFMAX(1, ff_norm_qscale(qp, p->qscale_type));
286 for (i = 0; i < count; i++) {
287 const int x1 = x + offset[i + count - 1][0];
288 const int y1 = y + offset[i + count - 1][1];
289 const int index = x1 + y1*linesize;
290 p->dct->get_pixels_unaligned(block, p->src + sample_bytes*index, sample_bytes*linesize);
292 p->requantize(block2, block, qp, p->dct->idct_permutation);
293 p->dct->idct(block2);
294 add_block(p->temp + index, linesize, block2);
298 if (sample_bytes == 1) {
299 p->store_slice(dst + (y - 8) * dst_linesize, p->temp + 8 + y*linesize,
300 dst_linesize, linesize, width,
301 FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
304 store_slice16_c((uint16_t*)(dst + (y - 8) * dst_linesize), p->temp + 8 + y*linesize,
305 dst_linesize/2, linesize, width,
306 FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
313 static int query_formats(AVFilterContext *ctx)
315 static const enum AVPixelFormat pix_fmts[] = {
316 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P,
317 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P,
318 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV440P,
319 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ422P,
320 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ440P,
321 AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV422P10,
322 AV_PIX_FMT_YUV420P10,
323 AV_PIX_FMT_YUV444P9, AV_PIX_FMT_YUV422P9,
332 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
334 return AVERROR(ENOMEM);
335 return ff_set_common_formats(ctx, fmts_list);
338 static int config_input(AVFilterLink *inlink)
340 SPPContext *s = inlink->dst->priv;
341 const int h = FFALIGN(inlink->h + 16, 16);
342 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
343 const int bps = desc->comp[0].depth;
345 av_opt_set_int(s->dct, "bits_per_sample", bps, 0);
346 avcodec_dct_init(s->dct);
351 s->hsub = desc->log2_chroma_w;
352 s->vsub = desc->log2_chroma_h;
353 s->temp_linesize = FFALIGN(inlink->w + 16, 16);
354 s->temp = av_malloc_array(s->temp_linesize, h * sizeof(*s->temp));
355 s->src = av_malloc_array(s->temp_linesize, h * sizeof(*s->src) * 2);
357 if (!s->temp || !s->src)
358 return AVERROR(ENOMEM);
362 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
364 AVFilterContext *ctx = inlink->dst;
365 SPPContext *s = ctx->priv;
366 AVFilterLink *outlink = ctx->outputs[0];
369 int8_t *qp_table = NULL;
370 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
371 const int depth = desc->comp[0].depth;
374 /* if we are not in a constant user quantizer mode and we don't want to use
375 * the quantizers from the B-frames (B-frames often have a higher QP), we
376 * need to save the qp table from the last non B-frame; this is what the
377 * following code block does */
378 if (!s->qp && (s->use_bframe_qp || in->pict_type != AV_PICTURE_TYPE_B)) {
379 ret = ff_qp_table_extract(in, &qp_table, &qp_stride, NULL, &s->qscale_type);
385 if (!s->use_bframe_qp && in->pict_type != AV_PICTURE_TYPE_B) {
386 av_freep(&s->non_b_qp_table);
387 s->non_b_qp_table = qp_table;
388 s->non_b_qp_stride = qp_stride;
392 if (s->log2_count && !ctx->is_disabled) {
393 if (!s->use_bframe_qp && s->non_b_qp_table) {
394 qp_table = s->non_b_qp_table;
395 qp_stride = s->non_b_qp_stride;
398 if (qp_table || s->qp) {
399 const int cw = AV_CEIL_RSHIFT(inlink->w, s->hsub);
400 const int ch = AV_CEIL_RSHIFT(inlink->h, s->vsub);
402 /* get a new frame if in-place is not possible or if the dimensions
403 * are not multiple of 8 */
404 if (!av_frame_is_writable(in) || (inlink->w & 7) || (inlink->h & 7)) {
405 const int aligned_w = FFALIGN(inlink->w, 8);
406 const int aligned_h = FFALIGN(inlink->h, 8);
408 out = ff_get_video_buffer(outlink, aligned_w, aligned_h);
411 ret = AVERROR(ENOMEM);
414 av_frame_copy_props(out, in);
415 out->width = in->width;
416 out->height = in->height;
419 filter(s, out->data[0], in->data[0], out->linesize[0], in->linesize[0], inlink->w, inlink->h, qp_table, qp_stride, 1, depth);
422 filter(s, out->data[1], in->data[1], out->linesize[1], in->linesize[1], cw, ch, qp_table, qp_stride, 0, depth);
423 filter(s, out->data[2], in->data[2], out->linesize[2], in->linesize[2], cw, ch, qp_table, qp_stride, 0, depth);
431 av_image_copy_plane(out->data[3], out->linesize[3],
432 in ->data[3], in ->linesize[3],
433 inlink->w, inlink->h);
436 ret = ff_filter_frame(outlink, out);
438 if (qp_table != s->non_b_qp_table)
443 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
444 char *res, int res_len, int flags)
446 SPPContext *s = ctx->priv;
448 if (!strcmp(cmd, "level") || !strcmp(cmd, "quality")) {
449 if (!strcmp(args, "max"))
450 s->log2_count = MAX_LEVEL;
452 s->log2_count = av_clip(strtol(args, NULL, 10), 0, MAX_LEVEL);
455 return AVERROR(ENOSYS);
458 static av_cold int init_dict(AVFilterContext *ctx, AVDictionary **opts)
460 SPPContext *s = ctx->priv;
463 s->dct = avcodec_dct_alloc();
465 return AVERROR(ENOMEM);
468 AVDictionaryEntry *e = NULL;
470 while ((e = av_dict_get(*opts, "", e, AV_DICT_IGNORE_SUFFIX))) {
471 if ((ret = av_opt_set(s->dct, e->key, e->value, 0)) < 0)
477 s->store_slice = store_slice_c;
479 case MODE_HARD: s->requantize = hardthresh_c; break;
480 case MODE_SOFT: s->requantize = softthresh_c; break;
485 static av_cold void uninit(AVFilterContext *ctx)
487 SPPContext *s = ctx->priv;
492 av_freep(&s->non_b_qp_table);
495 static const AVFilterPad spp_inputs[] = {
498 .type = AVMEDIA_TYPE_VIDEO,
499 .config_props = config_input,
500 .filter_frame = filter_frame,
505 static const AVFilterPad spp_outputs[] = {
508 .type = AVMEDIA_TYPE_VIDEO,
513 AVFilter ff_vf_spp = {
515 .description = NULL_IF_CONFIG_SMALL("Apply a simple post processing filter."),
516 .priv_size = sizeof(SPPContext),
517 .init_dict = init_dict,
519 .query_formats = query_formats,
520 .inputs = spp_inputs,
521 .outputs = spp_outputs,
522 .process_command = process_command,
523 .priv_class = &spp_class,
524 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,