Bug 520078 – Rotate brushes

2009-02-28  Sven Neumann  <sven@gimp.org>

	Bug 520078 – Rotate brushes

	* app/core/gimpbrush-transform.c: applied patch from Tal that
	implements bilinear interpolation for the brush transformations.


svn path=/trunk/; revision=28078
This commit is contained in:
Sven Neumann 2009-02-28 13:06:44 +00:00 committed by Sven Neumann
parent 0c17e790a7
commit a586ea03fa
2 changed files with 389 additions and 41 deletions

View File

@ -1,3 +1,10 @@
2009-02-28 Sven Neumann <sven@gimp.org>
Bug 520078 Rotate brushes
* app/core/gimpbrush-transform.c: applied patch from Tal that
implements bilinear interpolation for the brush transformations.
2009-02-28 Martin Nordholts <martinn@svn.gnome.org>
* app/actions/debug-commands.c: Properly show name of image graph

View File

@ -63,6 +63,27 @@ gimp_brush_real_transform_size (GimpBrush *brush,
gimp_brush_transform_bounding_box (brush, &matrix, &x, &y, width, height);
}
/*
* Transforms the brush mask with bilinear interpolation.
*
* Rather than calculating the inverse transform for each point in the
* transformed image, this algorithm uses the inverse transformed corner
* points of the destination image to work out the starting position in the
* source image and the U and V deltas in the source image space.
* It then uses a scan-line approach, looping through rows and colummns
* in the transformed (destination) image while walking along the corresponding
* rows and columns (named U and V) in the source image.
*
* The horizontal in destination space (transform result) is reverse transformed
* into source image space to get U.
* The vertical in destination space (transform result) is reverse transformed
* into source image space to get V.
*
* The strength of this particular algorithm is that calculation work should
* depend more upon the final transformed brush size rather than the input brush size.
*
* There are no floating point calculations in the inner loop for speed.
*/
TempBuf *
gimp_brush_real_transform_mask (GimpBrush *brush,
gdouble scale_x,
@ -78,6 +99,58 @@ gimp_brush_real_transform_mask (GimpBrush *brush,
gint dest_width;
gint dest_height;
gint x, y;
gdouble blx, brx, tlx, trx;
gdouble bly, bry, tly, try;
gdouble src_tl_to_tr_delta_x;
gdouble src_tl_to_tr_delta_y;
gdouble src_tl_to_bl_delta_x;
gdouble src_tl_to_bl_delta_y;
gint src_walk_ux;
gint src_walk_uy;
gint src_walk_vx;
gint src_walk_vy;
gint src_space_cur_pos_x;
gint src_space_cur_pos_y;
gint src_space_row_start_x;
gint src_space_row_start_y;
guchar *src_walker;
guchar *pixel_next;
guchar *pixel_below;
guchar *pixel_below_next;
gint opposite_x, distance_from_true_x;
gint opposite_y, distance_from_true_y;
gint src_height_times_int_multiple;
gint src_width_times_int_multiple;
/*
* tl, tr etc are used because it is easier to visualize top left, top right etc
* corners of the forward transformed source image rectangle.
*/
const gint fraction_bits = 8;
const gint int_multiple = pow(2,fraction_bits);
/* In inner loop's bilinear calculation, two numbers that were each previously multiplied by
* int_multiple are multiplied together.
* To get back the right result, the multiplication result must be
* divided *twice* by 2^fraction_bits, equivalent to
* bit shift right by 2 * fraction_bits
*/
const gint recovery_bits = 2 * fraction_bits;
/*
* example: suppose fraction_bits = 9
* a 9-bit mask looks like this: 0001 1111 1111
* and is given by: 2^fraction_bits - 1
* demonstration:
* 2^0 = 0000 0000 0001
* 2^1 = 0000 0000 0010
* :
* 2^8 = 0001 0000 0000
* 2^9 = 0010 0000 0000
* 2^9 - 1 = 0001 1111 1111
*/
const guint fraction_bitmask = pow(2, fraction_bits) - 1 ;
gimp_brush_transform_matrix (brush, scale_x, scale_y, angle, &matrix);
@ -92,45 +165,165 @@ gimp_brush_real_transform_mask (GimpBrush *brush,
gimp_matrix3_translate (&matrix, -x, -y);
gimp_matrix3_invert (&matrix);
result = temp_buf_new (dest_width, dest_height, 1, 0, 0, NULL);
result = temp_buf_new (dest_width, dest_height, 1, 0, 0, NULL); //3 instead of 1
dest = temp_buf_get_data (result);
src = temp_buf_get_data (brush->mask);
gimp_matrix3_transform_point (&matrix, 0, 0, &tlx, &tly);
gimp_matrix3_transform_point (&matrix, dest_width, 0, &trx, &try);
gimp_matrix3_transform_point (&matrix, 0, dest_height, &blx, &bly);
gimp_matrix3_transform_point (&matrix, dest_width, dest_height, &brx, &bry);
/* in image space, calc U (what was horizontal originally)
* note: double precision
*/
src_tl_to_tr_delta_x = trx - tlx;
src_tl_to_tr_delta_y = try - tly;
/* in image space, calc V (what was vertical originally)
* note: double precision
*/
src_tl_to_bl_delta_x = blx - tlx;
src_tl_to_bl_delta_y = bly - tly;
/* speed optimized, note conversion to int precision */
src_walk_ux = (gint) ((src_tl_to_tr_delta_x / dest_width)* int_multiple);
src_walk_uy = (gint) ((src_tl_to_tr_delta_y / dest_width)* int_multiple);
src_walk_vx = (gint) ((src_tl_to_bl_delta_x / dest_height)* int_multiple);
src_walk_vy = (gint) ((src_tl_to_bl_delta_y / dest_height)* int_multiple);
/* initialize current position in source space to the start position (tl)
* speed optimized, note conversion to int precision
*/
src_space_cur_pos_x = (gint) (tlx* int_multiple);
src_space_cur_pos_y = (gint) (tly* int_multiple);
src_space_row_start_x = (gint) (tlx* int_multiple);
src_space_row_start_y = (gint) (tly* int_multiple);
src_walker = src;
src_height_times_int_multiple = src_height << fraction_bits; /* mult by int_multiple */
src_width_times_int_multiple = src_width << fraction_bits; /* mult by int_multiple */
const gint src_heightm1_times_int_multiple = src_height_times_int_multiple - int_multiple;
const gint src_widthm1_times_int_multiple = src_width_times_int_multiple - int_multiple;
for (y = 0; y < dest_height; y++)
{
for (x = 0; x < dest_width; x++)
{
gdouble dx, dy;
gint ix, iy;
gimp_matrix3_transform_point (&matrix, x, y, &dx, &dy);
ix = ROUND (dx);
iy = ROUND (dy);
if (ix > 0 && ix < src_width &&
iy > 0 && iy < src_height)
{
*dest = src[iy * src_width + ix];
}
else
if (src_space_cur_pos_x > src_width_times_int_multiple ||
src_space_cur_pos_x < 0 ||
src_space_cur_pos_y > src_height_times_int_multiple ||
src_space_cur_pos_y < 0)
/* no corresponding pixel in source space */
{
*dest = 0;
/* dest[0] = 0;
dest[1] = 0;
dest[2] = 0;*/
}
else /* reverse transformed point hits source pixel */
{
src_walker = src
+ (src_space_cur_pos_y>>fraction_bits) * src_width
+ (src_space_cur_pos_x>>fraction_bits);
/* bottom right corner
* no pixel below, reuse current pixel instead
* no next pixel to the right so reuse current pixel instead
*/
if (src_space_cur_pos_y > (src_heightm1_times_int_multiple) &&
src_space_cur_pos_x > (src_widthm1_times_int_multiple) )
{
pixel_next = src_walker;
pixel_below = src_walker;
pixel_below_next = src_walker;
}
/* bottom edge pixel row, except rightmost corner
* no pixel below, reuse current pixel instead */
else if (src_space_cur_pos_y > (src_heightm1_times_int_multiple))
{
pixel_next = src_walker + 1;
pixel_below = src_walker;
pixel_below_next = src_walker + 1;
}
/* right edge pixel column, except bottom corner
* no next pixel to the right so reuse current pixel instead */
else if (src_space_cur_pos_x > (src_widthm1_times_int_multiple))
{
pixel_next = src_walker;
pixel_below = src_walker + src_width;
pixel_below_next = pixel_below;
}
/* neither on bottom edge nor on right edge */
else
{
pixel_next = src_walker + 1;
pixel_below = src_walker + src_width;
pixel_below_next = pixel_below + 1;
}
distance_from_true_x = src_space_cur_pos_x & fraction_bitmask;
distance_from_true_y = src_space_cur_pos_y & fraction_bitmask;
opposite_x = int_multiple - distance_from_true_x;
opposite_y = int_multiple - distance_from_true_y;
*dest= ( (src_walker[0] * opposite_x + pixel_next[0] * distance_from_true_x) * opposite_y +
(pixel_below[0] * opposite_x + pixel_below_next[0] *distance_from_true_x) * distance_from_true_y
) >> recovery_bits;
}
dest++;
}
}
src_space_cur_pos_x+=src_walk_ux;
src_space_cur_pos_y+=src_walk_uy;
dest ++;
} /* end for x */
src_space_row_start_x +=src_walk_vx;
src_space_row_start_y +=src_walk_vy;
src_space_cur_pos_x = src_space_row_start_x;
src_space_cur_pos_y = src_space_row_start_y;
} /* end for y */
return result;
}
/*
* Transforms the brush pixemap with bilinear interpolation.
*
* The algorithm used is exactly the same as for the brush mask
* (gimp_brush_real_transform_mask) except it accounts for 3 color channels
* instead of 1 greyscale channel.
*
* Rather than calculating the inverse transform for each point in the
* transformed image, this algorithm uses the inverse transformed corner
* points of the destination image to work out the starting position in the
* source image and the U and V deltas in the source image space.
* It then uses a scan-line approach, looping through rows and colummns
* in the transformed (destination) image while walking along the corresponding
* rows and columns (named U and V) in the source image.
*
* The horizontal in destination space (transform result) is reverse transformed
* into source image space to get U.
* The vertical in destination space (transform result) is reverse transformed
* into source image space to get V.
*
* The strength of this particular algorithm is that calculation work should
* depend more upon the final transformed brush size rather than the input brush size.
*
* There are no floating point calculations in the inner loop for speed.
*/
TempBuf *
gimp_brush_real_transform_pixmap (GimpBrush *brush,
gdouble scale_x,
gdouble scale_y,
gdouble angle)
gdouble scale_x,
gdouble scale_y,
gdouble angle)
{
TempBuf *result;
guchar *dest;
@ -141,6 +334,58 @@ gimp_brush_real_transform_pixmap (GimpBrush *brush,
gint dest_width;
gint dest_height;
gint x, y;
gdouble blx, brx, tlx, trx;
gdouble bly, bry, tly, try;
gdouble src_tl_to_tr_delta_x;
gdouble src_tl_to_tr_delta_y;
gdouble src_tl_to_bl_delta_x;
gdouble src_tl_to_bl_delta_y;
gint src_walk_ux;
gint src_walk_uy;
gint src_walk_vx;
gint src_walk_vy;
gint src_space_cur_pos_x;
gint src_space_cur_pos_y;
gint src_space_row_start_x;
gint src_space_row_start_y;
guchar *src_walker;
guchar *pixel_next;
guchar *pixel_below;
guchar *pixel_below_next;
gint opposite_x, distance_from_true_x;
gint opposite_y, distance_from_true_y;
gint src_height_times_int_multiple;
gint src_width_times_int_multiple;
/*
* tl, tr etc are used because it is easier to visualize top left, top right etc
* corners of the forward transformed source image rectangle.
*/
const gint fraction_bits = 8;
const gint int_multiple = pow(2,fraction_bits);
/* In inner loop's bilinear calculation, two numbers that were each previously multiplied by
* int_multiple are multiplied together.
* To get back the right result, the multiplication result must be
* divided *twice* by 2^fraction_bits, equivalent to
* bit shift right by 2 * fraction_bits
*/
const gint recovery_bits = 2 * fraction_bits;
/*
* example: suppose fraction_bits = 9
* a 9-bit mask looks like this: 0001 1111 1111
* and is given by: 2^fraction_bits - 1
* demonstration:
* 2^0 = 0000 0000 0001
* 2^1 = 0000 0000 0010
* :
* 2^8 = 0001 0000 0000
* 2^9 = 0010 0000 0000
* 2^9 - 1 = 0001 1111 1111
*/
const guint fraction_bitmask = pow(2, fraction_bits)- 1 ;
gimp_brush_transform_matrix (brush, scale_x, scale_y, angle, &matrix);
@ -160,37 +405,133 @@ gimp_brush_real_transform_pixmap (GimpBrush *brush,
dest = temp_buf_get_data (result);
src = temp_buf_get_data (brush->pixmap);
gimp_matrix3_transform_point (&matrix, 0, 0, &tlx, &tly);
gimp_matrix3_transform_point (&matrix, dest_width, 0, &trx, &try);
gimp_matrix3_transform_point (&matrix, 0, dest_height, &blx, &bly);
gimp_matrix3_transform_point (&matrix, dest_width, dest_height, &brx, &bry);
/* in image space, calc U (what was horizontal originally)
* note: double precision
*/
src_tl_to_tr_delta_x = trx - tlx;
src_tl_to_tr_delta_y = try - tly;
/* in image space, calc V (what was vertical originally)
* note: double precision
*/
src_tl_to_bl_delta_x = blx - tlx;
src_tl_to_bl_delta_y = bly - tly;
/* speed optimized, note conversion to int precision */
src_walk_ux = (gint) ((src_tl_to_tr_delta_x / dest_width)* int_multiple);
src_walk_uy = (gint) ((src_tl_to_tr_delta_y / dest_width)* int_multiple);
src_walk_vx = (gint) ((src_tl_to_bl_delta_x / dest_height)* int_multiple);
src_walk_vy = (gint) ((src_tl_to_bl_delta_y / dest_height)* int_multiple);
/* initialize current position in source space to the start position (tl)
* speed optimized, note conversion to int precision
*/
src_space_cur_pos_x = (gint) (tlx* int_multiple);
src_space_cur_pos_y = (gint) (tly* int_multiple);
src_space_row_start_x = (gint) (tlx* int_multiple);
src_space_row_start_y = (gint) (tly* int_multiple);
src_walker = src;
src_height_times_int_multiple = src_height << fraction_bits; /* mult by int_multiple */
src_width_times_int_multiple = src_width << fraction_bits; /* mult by int_multiple */
const gint src_heightm1_times_int_multiple = src_height_times_int_multiple - int_multiple;
const gint src_widthm1_times_int_multiple = src_width_times_int_multiple - int_multiple;
for (y = 0; y < dest_height; y++)
{
for (x = 0; x < dest_width; x++)
{
gdouble dx, dy;
gint ix, iy;
gimp_matrix3_transform_point (&matrix, x, y, &dx, &dy);
ix = ROUND (dx);
iy = ROUND (dy);
if (ix > 0 && ix < src_width &&
iy > 0 && iy < src_height)
{
const guchar *s = src + 3 * (iy * src_width + ix);
dest[0] = s[0];
dest[1] = s[1];
dest[2] = s[2];
}
else
if (src_space_cur_pos_x > src_width_times_int_multiple ||
src_space_cur_pos_x < 0 ||
src_space_cur_pos_y > src_height_times_int_multiple ||
src_space_cur_pos_y < 0)
/* no corresponding pixel in source space */
{
dest[0] = 0;
dest[1] = 0;
dest[2] = 0;
}
else /* reverse transformed point hits source pixel */
{
src_walker = src
+ 3 * (
(src_space_cur_pos_y >> fraction_bits) * src_width
+ (src_space_cur_pos_x >> fraction_bits));
/* bottom right corner
* no pixel below, reuse current pixel instead
* no next pixel to the right so reuse current pixel instead
*/
if (src_space_cur_pos_y > (src_heightm1_times_int_multiple) &&
src_space_cur_pos_x > (src_widthm1_times_int_multiple) )
{
pixel_next = src_walker;
pixel_below = src_walker;
pixel_below_next = src_walker;
}
/* bottom edge pixel row, except rightmost corner
* no pixel below, reuse current pixel instead */
else if (src_space_cur_pos_y > (src_heightm1_times_int_multiple))
{
pixel_next = src_walker + 3;
pixel_below = src_walker;
pixel_below_next = src_walker + 3;
}
/* right edge pixel column, except bottom corner
* no next pixel to the right so reuse current pixel instead */
else if (src_space_cur_pos_x > (src_widthm1_times_int_multiple))
{
pixel_next = src_walker;
pixel_below = src_walker + src_width * 3;
pixel_below_next = pixel_below;
}
/* neither on bottom edge nor on right edge */
else
{
pixel_next = src_walker + 3;
pixel_below = src_walker + src_width * 3;
pixel_below_next = pixel_below + 3;
}
distance_from_true_x = src_space_cur_pos_x & fraction_bitmask;
distance_from_true_y = src_space_cur_pos_y & fraction_bitmask;
opposite_x = int_multiple - distance_from_true_x;
opposite_y = int_multiple - distance_from_true_y;
dest[0] = ((src_walker[0] * opposite_x + pixel_next[0] * distance_from_true_x) * opposite_y +
(pixel_below[0] * opposite_x + pixel_below_next[0] *distance_from_true_x) * distance_from_true_y
) >> recovery_bits;
dest[1] = ((src_walker[1] * opposite_x + pixel_next[1] * distance_from_true_x) * opposite_y +
(pixel_below[1] * opposite_x + pixel_below_next[1] *distance_from_true_x) * distance_from_true_y
) >> recovery_bits;
dest[2] = ((src_walker[2] * opposite_x + pixel_next[2] * distance_from_true_x) * opposite_y +
(pixel_below[2] * opposite_x + pixel_below_next[2] *distance_from_true_x) * distance_from_true_y
) >> recovery_bits;
}
src_space_cur_pos_x += src_walk_ux;
src_space_cur_pos_y += src_walk_uy;
dest += 3;
}
}
} /* end for x */
src_space_row_start_x +=src_walk_vx;
src_space_row_start_y +=src_walk_vy;
src_space_cur_pos_x = src_space_row_start_x;
src_space_cur_pos_y = src_space_row_start_y;
} /* end for y */
return result;
}