sergiotarxz/libvips
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

fix default sharpen

Browse Source 
sharpen with sigma 0.5 was doing nothing, since the new int precision
rules meant that we generated a point-point convolution.

This patch increases the int precision for sharpen, so we now work with
sigma down to 0.5.

Also: restore input colourspace. Previously, the output image was always
LabS (the computation space for sharpen). Now, it transforms back to the
input space. This is more in line with how other operators work.

Thanks 2h4dl for pointing this out.

See https://github.com/libvips/pyvips/issues/123
This commit is contained in:
John Cupitt 2019-08-28 09:16:40 +01:00
parent 378537121f
commit 46212e92b1
2 changed files with 13 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ChangeLog
View File

@ -14,6 +14,8 @@
- clip coding and interpretation on vips image read
- check image bounds for GIF load
- prevent over-pre-shrink in thumbnail [kleisauke]
- fix sharpen with sigma 0.5 [2h4dl]
- sharpen restores input colourspace
24/5/19 started 8.8.1
- improve realpath() use on older libc

16
libvips/convolution/sharpen.c
View File

@ -39,6 +39,9 @@
* - swap "radius" for "sigma", allows finer control
* - allow a much greater range of parameters
* - move to defaults suitable for screen output
* 28/8/19
* - fix sigma 0.5 case (thanks 2h4dl)
* - restore input colourspace
*/
/*
@ -170,11 +173,12 @@ vips_sharpen_build( VipsObject *object )
{
VipsObjectClass *class = VIPS_OBJECT_GET_CLASS( object );
VipsSharpen *sharpen = (VipsSharpen *) object;
VipsImage **t = (VipsImage **) vips_object_local_array( object, 7 );
VipsImage **t = (VipsImage **) vips_object_local_array( object, 8 );
VipsImage **args = (VipsImage **) vips_object_local_array( object, 2 );
VipsImage *in;
int i;
VipsInterpretation old_interpretation;
VIPS_GATE_START( "vips_sharpen_build: build" );
@ -190,6 +194,7 @@ vips_sharpen_build( VipsObject *object )
in = sharpen->in;
old_interpretation = in->Type;
if( vips_colourspace( in, &t[0], VIPS_INTERPRETATION_LABS, NULL ) )
return( -1 );
in = t[0];
@ -199,10 +204,10 @@ vips_sharpen_build( VipsObject *object )
vips_check_format( class->nickname, in, VIPS_FORMAT_SHORT ) )
return( -1 );
/* Stop at 20% of max ... a bit mean. We always sharpen a short,
/* Stop at 10% of max ... a bit mean. We always sharpen a short,
* so there's no point using a float mask.
*/
if( vips_gaussmat( &t[1], sharpen->sigma, 0.2,
if( vips_gaussmat( &t[1], sharpen->sigma, 0.1,
"separable", TRUE,
"precision", VIPS_PRECISION_INTEGER,
NULL ) )
@ -284,10 +289,11 @@ vips_sharpen_build( VipsObject *object )
g_object_set( object, "out", vips_image_new(), NULL );
/* Reattach the rest.
/* Reattach the rest, back to the start colourspace.
*/
if( vips_bandjoin2( t[5], t[3], &t[6], NULL ) ||
vips_image_write( t[6], sharpen->out ) )
vips_colourspace( t[6], &t[7], old_interpretation, NULL ) ||
vips_image_write( t[7], sharpen->out ) )
return( -1 );
VIPS_GATE_STOP( "vips_sharpen_build: build" );