X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavfilter%2Fvf_nlmeans.c;h=dcb5a0395334565e798d730eebb1b932846df246;hb=0fef412dffb74fef3494f7fae0c138c32a444484;hp=547cb80acddff9a1ae8ddf2a5f297bb0df72ba0e;hpb=4278f79ef61058e3133214253a636113c18fb90a;p=ffmpeg

diff --git a/libavfilter/vf_nlmeans.c b/libavfilter/vf_nlmeans.c
index 547cb80acdd..dcb5a039533 100644
--- a/libavfilter/vf_nlmeans.c
+++ b/libavfilter/vf_nlmeans.c
@@ -43,9 +43,6 @@ struct weighted_avg {
     float sum;
 };
 
-#define WEIGHT_LUT_NBITS 9
-#define WEIGHT_LUT_SIZE  (1<<WEIGHT_LUT_NBITS)
-
 typedef struct NLMeansContext {
     const AVClass *class;
     int nb_planes;
@@ -62,8 +59,7 @@ typedef struct NLMeansContext {
     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
-    float weight_lut[WEIGHT_LUT_SIZE];          // lookup table mapping (scaled) patch differences to their associated weights
-    float pdiff_lut_scale;                      // scale factor for patch differences before looking into the LUT
+    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;
@@ -100,44 +96,6 @@ static int query_formats(AVFilterContext *ctx)
     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 uint32_t get_integral_patch_value(const uint32_t *ii, int ii_lz_32, int x, int y, int p)
-{
-    const uint32_t a = ii[(y - p - 1) * ii_lz_32 + (x - p - 1)];
-    const uint32_t b = ii[(y - p - 1) * ii_lz_32 + (x + p    )];
-    const uint32_t d = ii[(y + p    ) * ii_lz_32 + (x - p - 1)];
-    const uint32_t e = ii[(y + p    ) * ii_lz_32 + (x + p    )];
-    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
@@ -327,8 +285,8 @@ static int config_input(AVFilterLink *inlink)
     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"
@@ -393,19 +351,58 @@ static int nlmeans_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs
     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 uint32_t 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 unsigned weight_lut_idx = patch_diff_sq * s->pdiff_lut_scale;
-                const float 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;
 }
@@ -524,11 +521,12 @@ static av_cold int init(AVFilterContext *ctx)
     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");
@@ -556,6 +554,7 @@ static av_cold int init(AVFilterContext *ctx)
 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);
 }