[gegl] buffer: improve accuracy of sampler box filter

[N/A <ell src gnome org>]

commit 600cbc710f90ea2873d0a8ad638c3c52500cf267
Author: Ell <ell_se yahoo com>
Date:   Sat Mar 3 14:39:38 2018 -0500

    buffer: improve accuracy of sampler box filter
    
    Previously, we sampled at uniform intervals of the bounding box of
    the transformed area, which is inaccurate, as it may contain points
    outside the transformed area.  Furthermore, the divergence from the
    ideal result depended on the transformation, so the quality of the
    result changed as the transformation changed, even when its scale
    remained constant.
    
    Instead, sample at uniform intervals along the transformed grid,
    such that all sampled points are inside the transformed area.  This
    is more accurate, and more consistent.
    
    Update affected tests.
    
    Also, Replace g_clear_pointer() with g_clear_object() in last
    commit.

 gegl/buffer/gegl-sampler.c              |    2 +-
 gegl/buffer/gegl-sampler.h              |   98 ++++++++++++++++---------------
 tests/compositions/reference/clones.png |  Bin 723352 -> 723387 bytes
 3 files changed, 51 insertions(+), 49 deletions(-)
---
diff --git a/gegl/buffer/gegl-sampler.c b/gegl/buffer/gegl-sampler.c
index 71cae9d..cdd17c3 100644
--- a/gegl/buffer/gegl-sampler.c
+++ b/gegl/buffer/gegl-sampler.c
@@ -242,7 +242,7 @@ dispose (GObject *gobject)
   /* This call handles unreffing the buffer and disconnecting signals */
   set_buffer (sampler, NULL);
 
-  g_clear_pointer (&sampler->nearest_sampler, g_object_unref);
+  g_clear_object (&sampler->nearest_sampler);
 
   G_OBJECT_CLASS (gegl_sampler_parent_class)->dispose (gobject);
 }
diff --git a/gegl/buffer/gegl-sampler.h b/gegl/buffer/gegl-sampler.h
index b194f73..dd68ba7 100644
--- a/gegl/buffer/gegl-sampler.h
+++ b/gegl/buffer/gegl-sampler.h
@@ -231,62 +231,64 @@ _gegl_sampler_box_get (GeglSampler*    restrict  self,
   if (scale && fabs (gegl_matrix2_determinant (scale)) >= 4.0)
     {
       gfloat result[4] = {0,0,0,0};
-      const gdouble x1 = MIN (MIN (scale->coeff[0][0], scale->coeff[0][1]),
-                              MIN (0.0, scale->coeff[0][0] + scale->coeff[0][1]));
-      const gdouble y1 = MIN (MIN (scale->coeff[1][0], scale->coeff[1][1]),
-                              MIN (0.0, scale->coeff[1][0] + scale->coeff[1][1]));
-      const gdouble x2 = MAX (MAX (scale->coeff[0][0], scale->coeff[0][1]),
-                              MAX (0.0, scale->coeff[0][0] + scale->coeff[0][1]));
-      const gdouble y2 = MAX (MAX (scale->coeff[1][0], scale->coeff[1][1]),
-                              MAX (0.0, scale->coeff[1][0] + scale->coeff[1][1]));
-      const gdouble w = x2 - x1;
-      const gdouble h = y2 - y1;
-      const gint ix = floor (absolute_x - w / 2.0);
-      const gint iy = floor (absolute_y - h / 2.0);
-      const gint xx = ceil  (absolute_x + w / 2.0);
-      const gint yy = ceil  (absolute_y + h / 2.0);
-      int u, v;
-      int count = 0;
-      int hskip = xx - ix;
-      int vskip = yy - iy;
-
-      if (hskip > 0 && vskip > 0)
+      const gdouble u_norm         = sqrt (scale->coeff[0][0] * scale->coeff[0][0] +
+                                           scale->coeff[1][0] * scale->coeff[1][0]);
+      const gdouble v_norm         = sqrt (scale->coeff[0][1] * scale->coeff[0][1] +
+                                           scale->coeff[1][1] * scale->coeff[1][1]);
+      const gint    u_samples      = ceil (MIN (u_norm, n_samples));
+      const gint    v_samples      = ceil (MIN (v_norm, n_samples));
+      const gdouble u_samples_inv  = 1.0 / u_samples;
+      const gdouble v_samples_inv  = 1.0 / v_samples;
+      const gdouble uv_samples_inv = u_samples_inv * v_samples_inv;
+      const gdouble u_dx           = scale->coeff[0][0] * u_samples_inv;
+      const gdouble u_dy           = scale->coeff[1][0] * u_samples_inv;
+      const gdouble v_dx           = scale->coeff[0][1] * v_samples_inv;
+      const gdouble v_dy           = scale->coeff[1][1] * v_samples_inv;
+      gdouble       x0             = absolute_x - (scale->coeff[0][0] - u_dx +
+                                                   scale->coeff[0][1] - v_dx) /
+                                                  2.0;
+      gdouble       y0             = absolute_y - (scale->coeff[1][0] - u_dy +
+                                                   scale->coeff[1][1] - v_dy) /
+                                                  2.0;
+      gint          u;
+      gint          v;
+
+      if (! self->nearest_sampler)
         {
-          hskip /= n_samples;
-          vskip /= n_samples;
-
-          hskip += ! hskip;
-          vskip += ! vskip;
+          self->nearest_sampler = gegl_buffer_sampler_new (self->buffer,
+                                                           self->format,
+                                                           GEGL_SAMPLER_NEAREST);
+          self->nearest_sampler_get_fun =
+            gegl_sampler_get_fun (self->nearest_sampler);
+        }
 
-          if (! self->nearest_sampler)
-            {
-              self->nearest_sampler = gegl_buffer_sampler_new (self->buffer,
-                                                               self->format,
-                                                               GEGL_SAMPLER_NEAREST);
-              self->nearest_sampler_get_fun =
-                gegl_sampler_get_fun (self->nearest_sampler);
-            }
+      for (v = 0; v < v_samples; v++)
+        {
+          gdouble x = x0;
+          gdouble y = y0;
 
-          for (v = iy; v < yy; v += vskip)
+          for (u = 0; u < u_samples; u++)
             {
-              for (u = ix; u < xx; u += hskip)
-                {
-                  int c;
-                  gfloat input[4];
-                  self->nearest_sampler_get_fun (self->nearest_sampler,
-                                                 u, v, NULL, input, repeat_mode);
-                  for (c = 0; c < 4; c++)
-                    result[c] += input[c];
-                  count ++;
-                }
+              int c;
+              gfloat input[4];
+              self->nearest_sampler_get_fun (self->nearest_sampler,
+                                             x, y, NULL, input, repeat_mode);
+              for (c = 0; c < 4; c++)
+                result[c] += input[c];
+
+              x += u_dx;
+              y += u_dy;
             }
 
-          result[0] /= count;
-          result[1] /= count;
-          result[2] /= count;
-          result[3] /= count;
+          x0 += v_dx;
+          y0 += v_dy;
         }
 
+      result[0] *= uv_samples_inv;
+      result[1] *= uv_samples_inv;
+      result[2] *= uv_samples_inv;
+      result[3] *= uv_samples_inv;
+
       babl_process (self->fish, result, output, 1);
 
       return TRUE;
diff --git a/tests/compositions/reference/clones.png b/tests/compositions/reference/clones.png
index 43b45fe..942fc02 100644
Binary files a/tests/compositions/reference/clones.png and b/tests/compositions/reference/clones.png differ