swscale: add two spatially stable dithering methods

Both of these dithering methods are from http://pippin.gimp.org/a_dither/ for
GIF they can be considered better than bayer (provides more gray-levels), and
spatial stability - often more than twice as good compression and less visual
flicker than error diffusion methods (the methods also avoids error-shadow
artifacts of diffusion dithers).

These methods are similar to blue/green noise type dither masks; but are
simple enough to generate their mask on the fly. They are still research work
in progress; though more expensive to generate masks (which can be used in a
LUT) like 'void and cluster' and similar methods will yield superior results

doc/scaler.texi

@@ -112,6 +112,14 @@ bayer dither
 
 @item ed
 error diffusion dither
+
+@item a_dither
+arithmetic dither, based using addition
+
+@item x_dither
+arithmetic dither, based using xor (more random/less apparent patterning that
+a_dither).
+
 @end table
 
 @end table

libswscale/options.c

@@ -69,10 +69,12 @@ static const AVOption swscale_options[] = {
     { "dst_v_chr_pos",   "destination vertical chroma position in luma grid/256"  , OFFSET(dst_v_chr_pos), AV_OPT_TYPE_INT, { .i64 = -1            }, -1,      512,             VE },
     { "dst_h_chr_pos",   "destination horizontal chroma position in luma grid/256", OFFSET(dst_h_chr_pos), AV_OPT_TYPE_INT, { .i64 = -1            }, -1,      512,             VE },
 
-    { "sws_dither",      "set dithering algorithm",       OFFSET(dither),    AV_OPT_TYPE_INT,    { .i64  = SWS_DITHER_AUTO   }, 0,       NB_SWS_DITHER,  VE, "sws_dither" },
-    { "auto",            "leave choice to sws",           0,                 AV_OPT_TYPE_CONST,  { .i64  = SWS_DITHER_AUTO   }, INT_MIN, INT_MAX,        VE, "sws_dither" },
-    { "bayer",           "bayer dither",                  0,                 AV_OPT_TYPE_CONST,  { .i64  = SWS_DITHER_BAYER  }, INT_MIN, INT_MAX,        VE, "sws_dither" },
-    { "ed",              "error diffusion",               0,                 AV_OPT_TYPE_CONST,  { .i64  = SWS_DITHER_ED     }, INT_MIN, INT_MAX,        VE, "sws_dither" },
+    { "sws_dither",      "set dithering algorithm",       OFFSET(dither),    AV_OPT_TYPE_INT,    { .i64  = SWS_DITHER_AUTO    }, 0,       NB_SWS_DITHER,  VE, "sws_dither" },
+    { "auto",            "leave choice to sws",           0,                 AV_OPT_TYPE_CONST,  { .i64  = SWS_DITHER_AUTO    }, INT_MIN, INT_MAX,        VE, "sws_dither" },
+    { "bayer",           "bayer dither",                  0,                 AV_OPT_TYPE_CONST,  { .i64  = SWS_DITHER_BAYER   }, INT_MIN, INT_MAX,        VE, "sws_dither" },
+    { "ed",              "error diffusion",               0,                 AV_OPT_TYPE_CONST,  { .i64  = SWS_DITHER_ED      }, INT_MIN, INT_MAX,        VE, "sws_dither" },
+    { "a_dither",        "arithmetic addition dither",    0,                 AV_OPT_TYPE_CONST,  { .i64  = SWS_DITHER_A_DITHER}, INT_MIN, INT_MAX,        VE, "sws_dither" },
+    { "x_dither",        "arithmetic xor dither",         0,                 AV_OPT_TYPE_CONST,  { .i64  = SWS_DITHER_X_DITHER}, INT_MIN, INT_MAX,        VE, "sws_dither" },
 
     { NULL }
 };

libswscale/output.c

@@ -1508,24 +1508,71 @@ static av_always_inline void yuv2rgb_write_full(SwsContext *c,
     case AV_PIX_FMT_RGB8:
     {
         int r,g,b;
-        R >>= 22;
-        G >>= 22;
-        B >>= 22;
-        R += (7*err[0] + 1*c->dither_error[0][i] + 5*c->dither_error[0][i+1] + 3*c->dither_error[0][i+2])>>4;
-        G += (7*err[1] + 1*c->dither_error[1][i] + 5*c->dither_error[1][i+1] + 3*c->dither_error[1][i+2])>>4;
-        B += (7*err[2] + 1*c->dither_error[2][i] + 5*c->dither_error[2][i+1] + 3*c->dither_error[2][i+2])>>4;
-        c->dither_error[0][i] = err[0];
-        c->dither_error[1][i] = err[1];
-        c->dither_error[2][i] = err[2];
-        r = R >> (isrgb8 ? 5 : 7);
-        g = G >> (isrgb8 ? 5 : 6);
-        b = B >> (isrgb8 ? 6 : 7);
-        r = av_clip(r, 0, isrgb8 ? 7 : 1);
-        g = av_clip(g, 0, isrgb8 ? 7 : 3);
-        b = av_clip(b, 0, isrgb8 ? 3 : 1);
-        err[0] = R - r*(isrgb8 ? 36 : 255);
-        err[1] = G - g*(isrgb8 ? 36 : 85);
-        err[2] = B - b*(isrgb8 ? 85 : 255);
+
+        switch (c->dither) {
+        default:
+        case SWS_DITHER_AUTO:
+        case SWS_DITHER_ED:
+            R >>= 22;
+            G >>= 22;
+            B >>= 22;
+            R += (7*err[0] + 1*c->dither_error[0][i] + 5*c->dither_error[0][i+1] + 3*c->dither_error[0][i+2])>>4;
+            G += (7*err[1] + 1*c->dither_error[1][i] + 5*c->dither_error[1][i+1] + 3*c->dither_error[1][i+2])>>4;
+            B += (7*err[2] + 1*c->dither_error[2][i] + 5*c->dither_error[2][i+1] + 3*c->dither_error[2][i+2])>>4;
+            c->dither_error[0][i] = err[0];
+            c->dither_error[1][i] = err[1];
+            c->dither_error[2][i] = err[2];
+            r = R >> (isrgb8 ? 5 : 7);
+            g = G >> (isrgb8 ? 5 : 6);
+            b = B >> (isrgb8 ? 6 : 7);
+            r = av_clip(r, 0, isrgb8 ? 7 : 1);
+            g = av_clip(g, 0, isrgb8 ? 7 : 3);
+            b = av_clip(b, 0, isrgb8 ? 3 : 1);
+            err[0] = R - r*(isrgb8 ? 36 : 255);
+            err[1] = G - g*(isrgb8 ? 36 : 85);
+            err[2] = B - b*(isrgb8 ? 85 : 255);
+            break;
+        case SWS_DITHER_A_DITHER:
+            if (isrgb8) {
+  /* see http://pippin.gimp.org/a_dither/ for details/origin */
+#define A_DITHER(u,v)   (((((u)+((v)*236))*119)&0xff))
+                r = (((R >> 19) + A_DITHER(i,y)  -96)>>8);
+                g = (((G >> 19) + A_DITHER(i + 17,y) - 96)>>8);
+                b = (((B >> 20) + A_DITHER(i + 17*2,y) -96)>>8);
+                r = av_clip(r, 0, 7);
+                g = av_clip(g, 0, 7);
+                b = av_clip(b, 0, 3);
+            } else {
+                r = (((R >> 21) + A_DITHER(i,y)-256)>>8);
+                g = (((G >> 19) + A_DITHER(i + 17,y)-256)>>8);
+                b = (((B >> 21) + A_DITHER(i + 17*2,y)-256)>>8);
+                r = av_clip(r, 0, 1);
+                g = av_clip(g, 0, 3);
+                b = av_clip(b, 0, 1);
+            }
+            break;
+        case SWS_DITHER_X_DITHER:
+            if (isrgb8) {
+  /* see http://pippin.gimp.org/a_dither/ for details/origin */
+#define X_DITHER(u,v)   (((((u)^((v)*237))*181)&0x1ff)/2)
+                r = (((R >> 19) + X_DITHER(i,y) - 96)>>8);
+                g = (((G >> 19) + X_DITHER(i + 17,y) - 96)>>8);
+                b = (((B >> 20) + X_DITHER(i + 17*2,y) - 96)>>8);
+                r = av_clip(r, 0, 7);
+                g = av_clip(g, 0, 7);
+                b = av_clip(b, 0, 3);
+            } else {
+                r = (((R >> 21) + X_DITHER(i,y)-256)>>8);
+                g = (((G >> 19) + X_DITHER(i + 17,y)-256)>>8);
+                b = (((B >> 21) + X_DITHER(i + 17*2,y)-256)>>8);
+                r = av_clip(r, 0, 1);
+                g = av_clip(g, 0, 3);
+                b = av_clip(b, 0, 1);
+            }
+
+            break;
+        }
+
         if(target == AV_PIX_FMT_BGR4_BYTE) {
             dest[0] = r + 2*g + 8*b;
         } else if(target == AV_PIX_FMT_RGB4_BYTE) {

libswscale/swscale_internal.h

@@ -66,6 +66,8 @@ typedef enum SwsDither {
     SWS_DITHER_AUTO,
     SWS_DITHER_BAYER,
     SWS_DITHER_ED,
+    SWS_DITHER_A_DITHER,
+    SWS_DITHER_X_DITHER,
     NB_SWS_DITHER,
 } SwsDither;
 

libswscale/utils.c

@@ -1250,7 +1250,7 @@ av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
         if (c->dither == SWS_DITHER_AUTO)
             c->dither = (flags & SWS_FULL_CHR_H_INT) ? SWS_DITHER_ED : SWS_DITHER_BAYER;
         if (!(flags & SWS_FULL_CHR_H_INT)) {
-            if (c->dither == SWS_DITHER_ED) {
+            if (c->dither == SWS_DITHER_ED || c->dither == SWS_DITHER_A_DITHER || c->dither == SWS_DITHER_X_DITHER) {
                 av_log(c, AV_LOG_DEBUG,
                     "Desired dithering only supported in full chroma interpolation for destination format '%s'\n",
                     av_get_pix_fmt_name(dstFormat));