summaryrefslogtreecommitdiff
path: root/libavfilter/colorspacedsp_template.c
diff options
context:
space:
mode:
authorRonald S. Bultje <rsbultje@gmail.com>2016-05-03 13:53:17 -0400
committerRonald S. Bultje <rsbultje@gmail.com>2016-05-10 08:37:55 -0400
commit431117dcb55c9bd47febf9ee3379e14a5d6595c8 (patch)
treee4bb5bed1f77a604e4368fbe77267e4fbfb5b91b /libavfilter/colorspacedsp_template.c
parent9486de5a85f0c54f241849d6b22bee943da557d0 (diff)
downloadffmpeg-431117dcb55c9bd47febf9ee3379e14a5d6595c8.tar.gz
vf_colorspace: add floyd-steinberg dithering option to full conversion.
Diffstat (limited to 'libavfilter/colorspacedsp_template.c')
-rw-r--r--libavfilter/colorspacedsp_template.c128
1 files changed, 128 insertions, 0 deletions
diff --git a/libavfilter/colorspacedsp_template.c b/libavfilter/colorspacedsp_template.c
index f225391301..db4a8d2361 100644
--- a/libavfilter/colorspacedsp_template.c
+++ b/libavfilter/colorspacedsp_template.c
@@ -199,6 +199,134 @@ static void fn(rgb2yuv)(uint8_t *_yuv[3], ptrdiff_t yuv_stride[3],
}
}
+/* floyd-steinberg dithering - for any mid-top pixel A in a 3x2 block of pixels:
+ * 1 A 2
+ * 3 4 5
+ * the rounding error is distributed over the neighbouring pixels:
+ * 2: 7/16th, 3: 3/16th, 4: 5/16th and 5: 1/16th
+ */
+static void fn(rgb2yuv_fsb)(uint8_t *_yuv[3], ptrdiff_t yuv_stride[3],
+ int16_t *rgb[3], ptrdiff_t s,
+ int w, int h, const int16_t rgb2yuv_coeffs[3][3][8],
+ const int16_t yuv_offset[8],
+ int *rnd_scratch[3][2])
+{
+ pixel **yuv = (pixel **) _yuv;
+ pixel *yuv0 = yuv[0], *yuv1 = yuv[1], *yuv2 = yuv[2];
+ const int16_t *rgb0 = rgb[0], *rgb1 = rgb[1], *rgb2 = rgb[2];
+ int y, x;
+ const int sh = 29 - BIT_DEPTH;
+ const int rnd = 1 << (sh - 1);
+ int cry = rgb2yuv_coeffs[0][0][0];
+ int cgy = rgb2yuv_coeffs[0][1][0];
+ int cby = rgb2yuv_coeffs[0][2][0];
+ int cru = rgb2yuv_coeffs[1][0][0];
+ int cgu = rgb2yuv_coeffs[1][1][0];
+ int cburv = rgb2yuv_coeffs[1][2][0];
+ int cgv = rgb2yuv_coeffs[2][1][0];
+ int cbv = rgb2yuv_coeffs[2][2][0];
+ ptrdiff_t s0 = yuv_stride[0] / sizeof(pixel);
+ const int uv_offset = 128 << (BIT_DEPTH - 8);
+ unsigned mask = (1 << sh) - 1;
+
+ for (x = 0; x < w; x++) {
+ rnd_scratch[0][0][x] =
+ rnd_scratch[0][1][x] = rnd;
+ }
+ av_assert2(rgb2yuv_coeffs[1][2][0] == rgb2yuv_coeffs[2][0][0]);
+ w = AV_CEIL_RSHIFT(w, SS_W);
+ h = AV_CEIL_RSHIFT(h, SS_H);
+ for (x = 0; x < w; x++) {
+ rnd_scratch[1][0][x] =
+ rnd_scratch[1][1][x] =
+ rnd_scratch[2][0][x] =
+ rnd_scratch[2][1][x] = rnd;
+ }
+ for (y = 0; y < h; y++) {
+ for (x = 0; x < w; x++) {
+ int r00 = rgb0[x << SS_W], g00 = rgb1[x << SS_W], b00 = rgb2[x << SS_W];
+ int y00;
+#if SS_W == 1
+ int r01 = rgb0[x * 2 + 1], g01 = rgb1[x * 2 + 1], b01 = rgb2[x * 2 + 1];
+ int y01;
+#if SS_H == 1
+ int r10 = rgb0[x * 2 + 0 + s], g10 = rgb1[x * 2 + 0 + s], b10 = rgb2[x * 2 + 0 + s];
+ int r11 = rgb0[x * 2 + 1 + s], g11 = rgb1[x * 2 + 1 + s], b11 = rgb2[x * 2 + 1 + s];
+ int y10, y11;
+#endif
+#endif
+ int u, v, diff;
+
+ y00 = r00 * cry + g00 * cgy + b00 * cby + rnd_scratch[0][y & !SS_H][x << SS_W];
+ diff = (y00 & mask) - rnd;
+ yuv0[x << SS_W] = av_clip_pixel(yuv_offset[0] + (y00 >> sh));
+ rnd_scratch[0][ (y & !SS_H)][(x << SS_W) + 1] += (diff * 7 + 8) >> 4;
+ rnd_scratch[0][!(y & !SS_H)][(x << SS_W) - 1] += (diff * 3 + 8) >> 4;
+ rnd_scratch[0][!(y & !SS_H)][(x << SS_W) + 0] += (diff * 5 + 8) >> 4;
+ rnd_scratch[0][!(y & !SS_H)][(x << SS_W) + 1] += (diff * 1 + 8) >> 4;
+ rnd_scratch[0][ (y & !SS_H)][(x << SS_W) + 0] = rnd;
+#if SS_W == 1
+ y01 = r01 * cry + g01 * cgy + b01 * cby + rnd_scratch[0][y & !SS_H][x * 2 + 1];
+ diff = (y01 & mask) - rnd;
+ yuv0[x * 2 + 1] = av_clip_pixel(yuv_offset[0] + (y01 >> sh));
+ rnd_scratch[0][ (y & !SS_H)][x * 2 + 2] += (diff * 7 + 8) >> 4;
+ rnd_scratch[0][!(y & !SS_H)][x * 2 + 0] += (diff * 3 + 8) >> 4;
+ rnd_scratch[0][!(y & !SS_H)][x * 2 + 1] += (diff * 5 + 8) >> 4;
+ rnd_scratch[0][!(y & !SS_H)][x * 2 + 2] += (diff * 1 + 8) >> 4;
+ rnd_scratch[0][ (y & !SS_H)][x * 2 + 1] = rnd;
+#if SS_H == 1
+ y10 = r10 * cry + g10 * cgy + b10 * cby + rnd_scratch[0][1][x * 2 + 0];
+ diff = (y10 & mask) - rnd;
+ yuv0[x * 2 + 0 + s0] = av_clip_pixel(yuv_offset[0] + (y10 >> sh));
+ rnd_scratch[0][1][x * 2 + 1] += (diff * 7 + 8) >> 4;
+ rnd_scratch[0][0][x * 2 - 1] += (diff * 3 + 8) >> 4;
+ rnd_scratch[0][0][x * 2 + 0] += (diff * 5 + 8) >> 4;
+ rnd_scratch[0][0][x * 2 + 1] += (diff * 1 + 8) >> 4;
+ rnd_scratch[0][1][x * 2 + 0] = rnd;
+
+ y11 = r11 * cry + g11 * cgy + b11 * cby + rnd_scratch[0][1][x * 2 + 1];
+ diff = (y11 & mask) - rnd;
+ yuv0[x * 2 + 1 + s0] = av_clip_pixel(yuv_offset[0] + (y11 >> sh));
+ rnd_scratch[0][1][x * 2 + 2] += (diff * 7 + 8) >> 4;
+ rnd_scratch[0][0][x * 2 + 0] += (diff * 3 + 8) >> 4;
+ rnd_scratch[0][0][x * 2 + 1] += (diff * 5 + 8) >> 4;
+ rnd_scratch[0][0][x * 2 + 2] += (diff * 1 + 8) >> 4;
+ rnd_scratch[0][1][x * 2 + 1] = rnd;
+#endif
+#endif
+
+ u = avg(r00, r01, r10, r11) * cru +
+ avg(g00, g01, g10, g11) * cgu +
+ avg(b00, b01, b10, b11) * cburv + rnd_scratch[1][y & 1][x];
+ diff = (u & mask) - rnd;
+ yuv1[x] = av_clip_pixel(uv_offset + (u >> sh));
+ rnd_scratch[1][ (y & 1)][x + 1] += (diff * 7 + 8) >> 4;
+ rnd_scratch[1][!(y & 1)][x - 1] += (diff * 3 + 8) >> 4;
+ rnd_scratch[1][!(y & 1)][x + 0] += (diff * 5 + 8) >> 4;
+ rnd_scratch[1][!(y & 1)][x + 1] += (diff * 1 + 8) >> 4;
+ rnd_scratch[1][ (y & 1)][x + 0] = rnd;
+
+ v = avg(r00, r01, r10, r11) * cburv +
+ avg(g00, g01, g10, g11) * cgv +
+ avg(b00, b01, b10, b11) * cbv + rnd_scratch[2][y & 1][x];
+ diff = (v & mask) - rnd;
+ yuv2[x] = av_clip_pixel(uv_offset + (v >> sh));
+ rnd_scratch[2][ (y & 1)][x + 1] += (diff * 7 + 8) >> 4;
+ rnd_scratch[2][!(y & 1)][x - 1] += (diff * 3 + 8) >> 4;
+ rnd_scratch[2][!(y & 1)][x + 0] += (diff * 5 + 8) >> 4;
+ rnd_scratch[2][!(y & 1)][x + 1] += (diff * 1 + 8) >> 4;
+ rnd_scratch[2][ (y & 1)][x + 0] = rnd;
+ }
+
+ yuv0 += s0 * (1 << SS_H);
+ yuv1 += yuv_stride[1] / sizeof(pixel);
+ yuv2 += yuv_stride[2] / sizeof(pixel);
+ rgb0 += s * (1 << SS_H);
+ rgb1 += s * (1 << SS_H);
+ rgb2 += s * (1 << SS_H);
+ }
+}
+
#undef IN_BIT_DEPTH
#undef OUT_BIT_DEPTH
#define OUT_BIT_DEPTH BIT_DEPTH