/**
* @todo
- * - SIMD for final weighted averaging
* - better automatic defaults? see "Parameters" @ http://www.ipol.im/pub/art/2011/bcm_nlm/
* - temporal support (probably doesn't need any displacement according to
* "Denoising image sequences does not require motion estimation")
#include "video.h"
struct weighted_avg {
- double total_weight;
- double sum;
+ float total_weight;
+ float sum;
};
-#define WEIGHT_LUT_NBITS 9
-#define WEIGHT_LUT_SIZE (1<<WEIGHT_LUT_NBITS)
-
typedef struct NLMeansContext {
const AVClass *class;
int nb_planes;
ptrdiff_t ii_lz_32; // linesize in 32-bit units of the integral image
struct weighted_avg *wa; // weighted average of every pixel
ptrdiff_t wa_linesize; // linesize for wa in struct size unit
- double weight_lut[WEIGHT_LUT_SIZE]; // lookup table mapping (scaled) patch differences to their associated weights
- double pdiff_lut_scale; // scale factor for patch differences before looking into the LUT
- int max_meaningful_diff; // maximum difference considered (if the patch difference is too high we ignore the pixel)
+ float *weight_lut; // lookup table mapping (scaled) patch differences to their associated weights
+ uint32_t max_meaningful_diff; // maximum difference considered (if the patch difference is too high we ignore the pixel)
NLMeansDSPContext dsp;
} NLMeansContext;
return ff_set_common_formats(ctx, fmts_list);
}
-/*
- * M is a discrete map where every entry contains the sum of all the entries
- * in the rectangle from the top-left origin of M to its coordinate. In the
- * following schema, "i" contains the sum of the whole map:
- *
- * M = +----------+-----------------+----+
- * | | | |
- * | | | |
- * | a| b| c|
- * +----------+-----------------+----+
- * | | | |
- * | | | |
- * | | X | |
- * | | | |
- * | d| e| f|
- * +----------+-----------------+----+
- * | | | |
- * | g| h| i|
- * +----------+-----------------+----+
- *
- * The sum of the X box can be calculated with:
- * X = e-d-b+a
- *
- * See https://en.wikipedia.org/wiki/Summed_area_table
- *
- * The compute*_ssd functions compute the integral image M where every entry
- * contains the sum of the squared difference of every corresponding pixels of
- * two input planes of the same size as M.
- */
-static inline int get_integral_patch_value(const uint32_t *ii, int ii_lz_32, int x, int y, int p)
-{
- const int e = ii[(y + p ) * ii_lz_32 + (x + p )];
- const int d = ii[(y + p ) * ii_lz_32 + (x - p - 1)];
- const int b = ii[(y - p - 1) * ii_lz_32 + (x + p )];
- const int a = ii[(y - p - 1) * ii_lz_32 + (x - p - 1)];
- return e - d - b + a;
-}
-
/**
* Compute squared difference of the safe area (the zone where s1 and s2
* overlap). It is likely the largest integral zone, so it is interesting to do
* function, we do not need any clipping here.
*
* The line above dst and the column to its left are always readable.
- *
- * This C version computes the SSD integral image using a scalar accumulator,
- * while for SIMD implementation it is likely more interesting to use the
- * two-loops algorithm variant.
*/
static void compute_safe_ssd_integral_image_c(uint32_t *dst, ptrdiff_t dst_linesize_32,
const uint8_t *s1, ptrdiff_t linesize1,
int w, int h)
{
int x, y;
+ const uint32_t *dst_top = dst - dst_linesize_32;
/* SIMD-friendly assumptions allowed here */
av_assert2(!(w & 0xf) && w >= 16 && h >= 1);
for (y = 0; y < h; y++) {
- uint32_t acc = dst[-1] - dst[-dst_linesize_32 - 1];
-
- for (x = 0; x < w; x++) {
- const int d = s1[x] - s2[x];
- acc += d * d;
- dst[x] = dst[-dst_linesize_32 + x] + acc;
+ for (x = 0; x < w; x += 4) {
+ const int d0 = s1[x ] - s2[x ];
+ const int d1 = s1[x + 1] - s2[x + 1];
+ const int d2 = s1[x + 2] - s2[x + 2];
+ const int d3 = s1[x + 3] - s2[x + 3];
+
+ dst[x ] = dst_top[x ] - dst_top[x - 1] + d0*d0;
+ dst[x + 1] = dst_top[x + 1] - dst_top[x ] + d1*d1;
+ dst[x + 2] = dst_top[x + 2] - dst_top[x + 1] + d2*d2;
+ dst[x + 3] = dst_top[x + 3] - dst_top[x + 2] + d3*d3;
+
+ dst[x ] += dst[x - 1];
+ dst[x + 1] += dst[x ];
+ dst[x + 2] += dst[x + 1];
+ dst[x + 3] += dst[x + 2];
}
s1 += linesize1;
s2 += linesize2;
dst += dst_linesize_32;
+ dst_top += dst_linesize_32;
}
}
const int e = FFMAX(s->research_hsize, s->research_hsize_uv)
+ FFMAX(s->patch_hsize, s->patch_hsize_uv);
- s->chroma_w = FF_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
- s->chroma_h = FF_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
+ s->chroma_w = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
+ s->chroma_h = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
s->nb_planes = av_pix_fmt_count_planes(inlink->format);
/* Allocate the integral image with extra edges of thickness "e"
const int slice_end = (process_h * (jobnr+1)) / nb_jobs;
const int starty = td->starty + slice_start;
const int endy = td->starty + slice_end;
+ const int p = td->p;
+ const uint32_t *ii = td->ii_start + (starty - p - 1) * s->ii_lz_32 - p - 1;
+ const int dist_b = 2*p + 1;
+ const int dist_d = dist_b * s->ii_lz_32;
+ const int dist_e = dist_d + dist_b;
for (y = starty; y < endy; y++) {
const uint8_t *src = td->src + y*src_linesize;
struct weighted_avg *wa = s->wa + y*s->wa_linesize;
for (x = td->startx; x < td->endx; x++) {
- const int patch_diff_sq = get_integral_patch_value(td->ii_start, s->ii_lz_32, x, y, td->p);
+ /*
+ * M is a discrete map where every entry contains the sum of all the entries
+ * in the rectangle from the top-left origin of M to its coordinate. In the
+ * following schema, "i" contains the sum of the whole map:
+ *
+ * M = +----------+-----------------+----+
+ * | | | |
+ * | | | |
+ * | a| b| c|
+ * +----------+-----------------+----+
+ * | | | |
+ * | | | |
+ * | | X | |
+ * | | | |
+ * | d| e| f|
+ * +----------+-----------------+----+
+ * | | | |
+ * | g| h| i|
+ * +----------+-----------------+----+
+ *
+ * The sum of the X box can be calculated with:
+ * X = e-d-b+a
+ *
+ * See https://en.wikipedia.org/wiki/Summed_area_table
+ *
+ * The compute*_ssd functions compute the integral image M where every entry
+ * contains the sum of the squared difference of every corresponding pixels of
+ * two input planes of the same size as M.
+ */
+ const uint32_t a = ii[x];
+ const uint32_t b = ii[x + dist_b];
+ const uint32_t d = ii[x + dist_d];
+ const uint32_t e = ii[x + dist_e];
+ const uint32_t patch_diff_sq = e - d - b + a;
+
if (patch_diff_sq < s->max_meaningful_diff) {
- const int weight_lut_idx = patch_diff_sq * s->pdiff_lut_scale;
- const double weight = s->weight_lut[weight_lut_idx]; // exp(-patch_diff_sq * s->pdiff_scale)
+ const float weight = s->weight_lut[patch_diff_sq]; // exp(-patch_diff_sq * s->pdiff_scale)
wa[x].total_weight += weight;
wa[x].sum += weight * src[x];
}
}
+ ii += s->ii_lz_32;
}
return 0;
}
+static void weight_averages(uint8_t *dst, ptrdiff_t dst_linesize,
+ const uint8_t *src, ptrdiff_t src_linesize,
+ struct weighted_avg *wa, ptrdiff_t wa_linesize,
+ int w, int h)
+{
+ int x, y;
+
+ for (y = 0; y < h; y++) {
+ for (x = 0; x < w; x++) {
+ // Also weight the centered pixel
+ wa[x].total_weight += 1.f;
+ wa[x].sum += 1.f * src[x];
+ dst[x] = av_clip_uint8(wa[x].sum / wa[x].total_weight + 0.5f);
+ }
+ dst += dst_linesize;
+ src += src_linesize;
+ wa += wa_linesize;
+ }
+}
+
static int nlmeans_plane(AVFilterContext *ctx, int w, int h, int p, int r,
uint8_t *dst, ptrdiff_t dst_linesize,
const uint8_t *src, ptrdiff_t src_linesize)
{
- int x, y;
int offx, offy;
NLMeansContext *s = ctx->priv;
/* patches center points cover the whole research window so the patches
}
}
}
- for (y = 0; y < h; y++) {
- for (x = 0; x < w; x++) {
- struct weighted_avg *wa = &s->wa[y*s->wa_linesize + x];
- // Also weight the centered pixel
- wa->total_weight += 1.0;
- wa->sum += 1.0 * src[y*src_linesize + x];
+ weight_averages(dst, dst_linesize, src, src_linesize,
+ s->wa, s->wa_linesize, w, h);
- dst[y*dst_linesize + x] = av_clip_uint8(wa->sum / wa->total_weight);
- }
- }
return 0;
}
const double h = s->sigma * 10.;
s->pdiff_scale = 1. / (h * h);
- s->max_meaningful_diff = -log(1/255.) / s->pdiff_scale;
- s->pdiff_lut_scale = 1./s->max_meaningful_diff * WEIGHT_LUT_SIZE;
- av_assert0((s->max_meaningful_diff - 1) * s->pdiff_lut_scale < FF_ARRAY_ELEMS(s->weight_lut));
- for (i = 0; i < WEIGHT_LUT_SIZE; i++)
- s->weight_lut[i] = exp(-i / s->pdiff_lut_scale * s->pdiff_scale);
+ s->max_meaningful_diff = log(255.) / s->pdiff_scale;
+ s->weight_lut = av_calloc(s->max_meaningful_diff, sizeof(*s->weight_lut));
+ if (!s->weight_lut)
+ return AVERROR(ENOMEM);
+ for (i = 0; i < s->max_meaningful_diff; i++)
+ s->weight_lut[i] = exp(-i * s->pdiff_scale);
CHECK_ODD_FIELD(research_size, "Luma research window");
CHECK_ODD_FIELD(patch_size, "Luma patch");
static av_cold void uninit(AVFilterContext *ctx)
{
NLMeansContext *s = ctx->priv;
+ av_freep(&s->weight_lut);
av_freep(&s->ii_orig);
av_freep(&s->wa);
}
projects / ffmpeg / blobdiff