Code

Fixed some bad math in pixblock-scaler. Seems to fix shifted blurs
authorkiirala <kiirala@users.sourceforge.net>
Sat, 28 Mar 2009 11:35:16 +0000 (11:35 +0000)
committerkiirala <kiirala@users.sourceforge.net>
Sat, 28 Mar 2009 11:35:16 +0000 (11:35 +0000)
src/display/pixblock-scaler.cpp

index 511104c43a04c98e05306febca8cca6890245e7e..7673349353061852d1693863f6ef7f5fe957ec2f 100644 (file)
@@ -22,57 +22,32 @@ using std::floor;
 namespace NR {
 
 struct RGBA {
-    int r, g, b, a;
+    double r, g, b, a;
 };
 
 /** Calculates cubically interpolated value of the four given pixel values.
- * The pixel values should be from four vertically adjacent pixels.
- * If we are calculating a pixel, whose y-coordinate in source image is
- * i, these pixel values a, b, c and d should come from lines
- * floor(i) - 1, floor(i), floor(i) + 1, floor(i) + 2, respectively.
- * Parameter len should be set to i.
- * Returns the interpolated value in fixed point format with 8 bit
- * decimal part. (24.8 assuming 32-bit int)
+ * The pixel values should be from four adjacent pixels in source image or
+ * four adjacent interpolated values. len should be the x- or y-coordinate
+ * (depending on interpolation direction) of the center of the target pixel
+ * in source image coordinates.
  */
 __attribute__ ((const))
-inline static int sampley(unsigned const char a, unsigned const char b,
-                   unsigned const char c, unsigned const char d,
-                   const double len)
+inline static double sample(double const a, double const b,
+                            double const c, double const d,
+                            double const len)
 {
-    double lenf = len - floor(len);
-    int sum = 0;
-    sum += (int)((((-1.0 / 3.0) * lenf + 4.0 / 5.0) * lenf - 7.0 / 15.0)
-                 * lenf * 256 * a);
-    sum += (int)((((lenf - 9.0 / 5.0) * lenf - 1.0 / 5.0) * lenf + 1.0)
-                 * 256 * b);
-    sum += (int)(((((1 - lenf) - 9.0 / 5.0) * (1 - lenf) - 1.0 / 5.0)
-                  * (1 - lenf) + 1.0) * 256 * c);
-    sum += (int)((((-1.0 / 3.0) * (1 - lenf) + 4.0 / 5.0) * (1 - lenf)
-                  - 7.0 / 15.0) * (1 - lenf) * 256 * d);
-    return sum;
-}
+    double lena = 1.5 + (len - round(len));
+    double lenb = 0.5 + (len - round(len));
+    double lenc = 0.5 - (len - round(len));
+    double lend = 1.5 - (len - round(len));
+    double const f = -0.5; // corresponds to cubic Hermite spline
+    double sum = 0;
+    sum += ((((f * lena) - 5.0 * f) * lena + 8.0 * f) * lena - 4 * f) * a;
+    sum += (((f + 2.0) * lenb - (f + 3.0)) * lenb * lenb + 1.0) * b;
+    sum += (((f + 2.0) * lenc - (f + 3.0)) * lenc * lenc + 1.0) * c;
+    sum += ((((f * lend) - 5.0 * f) * lend + 8.0 * f) * lend - 4 * f) * d;
 
-/** Calculates cubically interpolated value of the four given pixel values.
- * The pixel values should be interpolated values from sampley, from four
- * horizontally adjacent vertical lines. The parameters a, b, c and d
- * should be in fixed point format with 8-bit decimal part.
- * If we are calculating a pixel, whose x-coordinate in source image is
- * i, these vertical  lines from where a, b, c and d are calculated, should be
- * floor(i) - 1, floor(i), floor(i) + 1, floor(i) + 2, respectively.
- * Parameter len should be set to i.
- * Returns the interpolated value in 8-bit format, ready to be written
- * to output buffer.
- */
-inline static int samplex(const int a, const int b, const int c, const int d, const double len) {
-    double lenf = len - floor(len);
-    int sum = 0;
-    sum += (int)(a * (((-1.0 / 3.0) * lenf + 4.0 / 5.0) * lenf - 7.0 / 15.0) * lenf);
-    sum += (int)(b * (((lenf - 9.0 / 5.0) * lenf - 1.0 / 5.0) * lenf + 1.0));
-    sum += (int)(c * ((((1 - lenf) - 9.0 / 5.0) * (1 - lenf) - 1.0 / 5.0) * (1 - lenf) + 1.0));
-    sum += (int)(d * (((-1.0 / 3.0) * (1 - lenf) + 4.0 / 5.0) * (1 - lenf) - 7.0 / 15.0) * (1 - lenf));
-    //if (sum < 0) sum = 0;
-    //if (sum > 255 * 256) sum = 255 * 256;
-    return sum / 256;
+    return sum;
 }
 
 /**
@@ -123,9 +98,10 @@ static void scale_bicubic_rgba(NRPixBlock *to, NRPixBlock *from)
         // which we should read
         int from_line[4];
         for (int i = 0 ; i < 4 ; i++) {
-            if ((int)floor(from_y) + i - 1 >= 0) {
-                if ((int)floor(from_y) + i - 1 < from_height) {
-                    from_line[i] = ((int)floor(from_y) + i - 1) * from->rs;
+            int fy_line = (int)round(from_y) + i - 2;
+            if (fy_line >= 0) {
+                if (fy_line < from_height) {
+                    from_line[i] = fy_line * from->rs;
                 } else {
                     from_line[i] = (from_height - 1) * from->rs;
                 }
@@ -140,7 +116,7 @@ static void scale_bicubic_rgba(NRPixBlock *to, NRPixBlock *from)
             double from_x = to_x * from_stepx + from_stepx / 2;
             RGBA line[4];
             for (int i = 0 ; i < 4 ; i++) {
-                int k = (int)floor(from_x) + i - 1;
+                int k = (int)round(from_x) + i - 2;
                 if (k < 0) k = 0;
                 if (k >= from_width) k = from_width - 1;
                 k *= 4;
@@ -148,36 +124,36 @@ static void scale_bicubic_rgba(NRPixBlock *to, NRPixBlock *from)
                 _check_index(from, from_line[1] + k, __LINE__);
                 _check_index(from, from_line[2] + k, __LINE__);
                 _check_index(from, from_line[3] + k, __LINE__);
-                line[i].r = sampley(NR_PIXBLOCK_PX(from)[from_line[0] + k],
-                                    NR_PIXBLOCK_PX(from)[from_line[1] + k],
-                                    NR_PIXBLOCK_PX(from)[from_line[2] + k],
-                                    NR_PIXBLOCK_PX(from)[from_line[3] + k],
-                                    from_y);
-                line[i].g = sampley(NR_PIXBLOCK_PX(from)[from_line[0] + k + 1],
-                                    NR_PIXBLOCK_PX(from)[from_line[1] + k + 1],
-                                    NR_PIXBLOCK_PX(from)[from_line[2] + k + 1],
-                                    NR_PIXBLOCK_PX(from)[from_line[3] + k + 1],
-                                    from_y);
-                line[i].b = sampley(NR_PIXBLOCK_PX(from)[from_line[0] + k + 2],
-                                    NR_PIXBLOCK_PX(from)[from_line[1] + k + 2],
-                                    NR_PIXBLOCK_PX(from)[from_line[2] + k + 2],
-                                    NR_PIXBLOCK_PX(from)[from_line[3] + k + 2],
-                                    from_y);
-                line[i].a = sampley(NR_PIXBLOCK_PX(from)[from_line[0] + k + 3],
-                                    NR_PIXBLOCK_PX(from)[from_line[1] + k + 3],
-                                    NR_PIXBLOCK_PX(from)[from_line[2] + k + 3],
-                                    NR_PIXBLOCK_PX(from)[from_line[3] + k + 3],
-                                    from_y);
+                line[i].r = sample(NR_PIXBLOCK_PX(from)[from_line[0] + k],
+                                   NR_PIXBLOCK_PX(from)[from_line[1] + k],
+                                   NR_PIXBLOCK_PX(from)[from_line[2] + k],
+                                   NR_PIXBLOCK_PX(from)[from_line[3] + k],
+                                   from_y);
+                line[i].g = sample(NR_PIXBLOCK_PX(from)[from_line[0] + k + 1],
+                                   NR_PIXBLOCK_PX(from)[from_line[1] + k + 1],
+                                   NR_PIXBLOCK_PX(from)[from_line[2] + k + 1],
+                                   NR_PIXBLOCK_PX(from)[from_line[3] + k + 1],
+                                   from_y);
+                line[i].b = sample(NR_PIXBLOCK_PX(from)[from_line[0] + k + 2],
+                                   NR_PIXBLOCK_PX(from)[from_line[1] + k + 2],
+                                   NR_PIXBLOCK_PX(from)[from_line[2] + k + 2],
+                                   NR_PIXBLOCK_PX(from)[from_line[3] + k + 2],
+                                   from_y);
+                line[i].a = sample(NR_PIXBLOCK_PX(from)[from_line[0] + k + 3],
+                                   NR_PIXBLOCK_PX(from)[from_line[1] + k + 3],
+                                   NR_PIXBLOCK_PX(from)[from_line[2] + k + 3],
+                                   NR_PIXBLOCK_PX(from)[from_line[3] + k + 3],
+                                   from_y);
             }
             RGBA result;
-            result.r = samplex(line[0].r, line[1].r, line[2].r, line[3].r,
-                               from_x);
-            result.g = samplex(line[0].g, line[1].g, line[2].g, line[3].g,
-                               from_x);
-            result.b = samplex(line[0].b, line[1].b, line[2].b, line[3].b,
-                               from_x);
-            result.a = samplex(line[0].a, line[1].a, line[2].a, line[3].a,
-                               from_x);
+            result.r = round(sample(line[0].r, line[1].r, line[2].r, line[3].r,
+                                    from_x));
+            result.g = round(sample(line[0].g, line[1].g, line[2].g, line[3].g,
+                                    from_x));
+            result.b = round(sample(line[0].b, line[1].b, line[2].b, line[3].b,
+                                    from_x));
+            result.a = round(sample(line[0].a, line[1].a, line[2].a, line[3].a,
+                                    from_x));
 
             _check_index(to, to_y * to->rs + to_x * 4, __LINE__);
 
@@ -188,17 +164,24 @@ static void scale_bicubic_rgba(NRPixBlock *to, NRPixBlock *from)
                  * make sure, we don't exceed 100% per colour channel with
                  * images that have premultiplied alpha */
 
-                result.a = clamp(result.a);
+                int const alpha = clamp((int)result.a);
 
-                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4] = clamp_alpha(result.r, result.a);
-                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 1] = clamp_alpha(result.g, result.a);
-                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 2] = clamp_alpha(result.b, result.a);
-                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 3] = result.a;
+                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4]
+                    = clamp_alpha((int)result.r, alpha);
+                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 1]
+                    = clamp_alpha((int)result.g, alpha);
+                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 2]
+                    = clamp_alpha((int)result.b, alpha);
+                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 3] = alpha;
             } else {
-                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4] = clamp(result.r);
-                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 1] = clamp(result.g);
-                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 2] = clamp(result.b);
-                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 3] = clamp(result.a);
+                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4]
+                    = clamp((int)result.r);
+                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 1]
+                    = clamp((int)result.g);
+                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 2]
+                    = clamp((int)result.b);
+                NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x * 4 + 3]
+                    = clamp((int)result.a);
             }
         }
     }
@@ -234,9 +217,10 @@ void scale_bicubic_alpha(NRPixBlock *to, NRPixBlock *from)
         // which we should read
         int from_line[4];
         for (int i = 0 ; i < 4 ; i++) {
-            if ((int)floor(from_y) + i - 1 >= 0) {
-                if ((int)floor(from_y) + i - 1 < from_height) {
-                    from_line[i] = ((int)floor(from_y) + i - 1) * from->rs;
+            int fy_line = (int)round(from_y) + i - 2;
+            if (fy_line >= 0) {
+                if (fy_line < from_height) {
+                    from_line[i] = fy_line * from->rs;
                 } else {
                     from_line[i] = (from_height - 1) * from->rs;
                 }
@@ -249,28 +233,29 @@ void scale_bicubic_alpha(NRPixBlock *to, NRPixBlock *from)
         // bicubic interpolation and set the pixel value in destination image
         for (int to_x = 0 ; to_x < to_width ; to_x++) {
             double from_x = to_x * from_stepx + from_stepx / 2;
-            int line[4];
+            double line[4];
             for (int i = 0 ; i < 4 ; i++) {
-                int k = (int)floor(from_x) + i - 1;
+                int k = (int)round(from_x) + i - 2;
                 if (k < 0) k = 0;
                 if (k >= from_width) k = from_width - 1;
                 _check_index(from, from_line[0] + k, __LINE__);
                 _check_index(from, from_line[1] + k, __LINE__);
                 _check_index(from, from_line[2] + k, __LINE__);
                 _check_index(from, from_line[3] + k, __LINE__);
-                line[i] = sampley(NR_PIXBLOCK_PX(from)[from_line[0] + k],
-                                  NR_PIXBLOCK_PX(from)[from_line[1] + k],
-                                  NR_PIXBLOCK_PX(from)[from_line[2] + k],
-                                  NR_PIXBLOCK_PX(from)[from_line[3] + k],
-                                  from_y);
+                line[i] = sample(NR_PIXBLOCK_PX(from)[from_line[0] + k],
+                                 NR_PIXBLOCK_PX(from)[from_line[1] + k],
+                                 NR_PIXBLOCK_PX(from)[from_line[2] + k],
+                                 NR_PIXBLOCK_PX(from)[from_line[3] + k],
+                                 from_y);
             }
             int result;
-            result = samplex(line[0], line[1], line[2], line[3],
-                             from_x);
+            result = (int)round(sample(line[0], line[1], line[2], line[3],
+                                       from_x));
 
             _check_index(to, to_y * to->rs + to_x, __LINE__);
 
-            NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x] = Inkscape::Filters::clamp(result);
+            NR_PIXBLOCK_PX(to)[to_y * to->rs + to_x]
+                = Inkscape::Filters::clamp(result);
         }
     }
 }