libvips/libvips/conversion/smartcrop.c

516 lines
13 KiB
C

/* crop an image down to a specified size by removing boring parts
*
* Adapted from sharp's smartcrop feature, with kind permission.
*
* 1/3/17
* - first version, from sharp
* 14/3/17
* - revised attention smartcrop
* 8/6/17
* - revised again
* 15/9/18 lovell
* - move shrink to start of processing
* 22/9/18 jcupitt
* - add low and high
* 19/3/20 jcupitt
* - add all
* 26/11/22 ejoebstl
* - expose location of interest when using attention based cropping
*/
/*
This file is part of VIPS.
VIPS is free software; you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
02110-1301 USA
*/
/*
These files are distributed with VIPS - http://www.vips.ecs.soton.ac.uk
*/
/*
#define VIPS_DEBUG
*/
#ifdef HAVE_CONFIG_H
#include <config.h>
#endif /*HAVE_CONFIG_H*/
#include <glib/gi18n-lib.h>
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <vips/vips.h>
#include <vips/debug.h>
#include "pconversion.h"
#include "bandary.h"
typedef struct _VipsSmartcrop {
VipsConversion parent_instance;
VipsImage *in;
int width;
int height;
VipsInteresting interesting;
int attention_x;
int attention_y;
} VipsSmartcrop;
typedef VipsConversionClass VipsSmartcropClass;
G_DEFINE_TYPE( VipsSmartcrop, vips_smartcrop, VIPS_TYPE_CONVERSION );
static int
vips_smartcrop_score( VipsSmartcrop *smartcrop, VipsImage *in,
int left, int top, int width, int height, double *score )
{
VipsImage **t = (VipsImage **)
vips_object_local_array( VIPS_OBJECT( smartcrop ), 2 );
if( vips_extract_area( in, &t[0], left, top, width, height, NULL ) ||
vips_hist_find( t[0], &t[1], NULL ) ||
vips_hist_entropy( t[1], score, NULL ) )
return( -1 );
return( 0 );
}
/* Entropy-style smartcrop. Repeatedly discard low interest areas. This should
* be faster for very large images.
*/
static int
vips_smartcrop_entropy( VipsSmartcrop *smartcrop,
VipsImage *in, int *left, int *top )
{
int max_slice_size;
int width;
int height;
*left = 0;
*top = 0;
width = in->Xsize;
height = in->Ysize;
/* How much do we trim by each iteration? Aim for 8 steps in the axis
* that needs trimming most.
*/
max_slice_size = VIPS_MAX(
ceil( (width - smartcrop->width) / 8.0 ),
ceil( (height - smartcrop->height) / 8.0 ) );
/* Repeatedly take a slice off width and height until we
* reach the target.
*/
while( width > smartcrop->width ||
height > smartcrop->height ) {
const int slice_width =
VIPS_MIN( width - smartcrop->width, max_slice_size );
const int slice_height =
VIPS_MIN( height - smartcrop->height, max_slice_size );
if( slice_width > 0 ) {
double left_score;
double right_score;
if( vips_smartcrop_score( smartcrop, in,
*left, *top,
slice_width, height, &left_score ) )
return( -1 );
if( vips_smartcrop_score( smartcrop, in,
*left + width - slice_width, *top,
slice_width, height, &right_score ) )
return( -1 );
width -= slice_width;
if( left_score < right_score )
*left += slice_width;
}
if( slice_height > 0 ) {
double top_score;
double bottom_score;
if( vips_smartcrop_score( smartcrop, in,
*left, *top,
width, slice_height, &top_score ) )
return( -1 );
if( vips_smartcrop_score( smartcrop, in,
*left, *top + height - slice_height,
width, slice_height, &bottom_score ) )
return( -1 );
height -= slice_height;
if( top_score < bottom_score )
*top += slice_height;
}
}
return( 0 );
}
/* Calculate sqrt(b1^2 + b2^2 ...)
*/
static int
pythagoras( VipsSmartcrop *smartcrop, VipsImage *in, VipsImage **out )
{
VipsImage **t = (VipsImage **)
vips_object_local_array( VIPS_OBJECT( smartcrop ),
2 * in->Bands + 1 );
int i;
for( i = 0; i < in->Bands; i++ )
if( vips_extract_band( in, &t[i], i, NULL ) )
return( -1 );
for( i = 0; i < in->Bands; i++ )
if( vips_multiply( t[i], t[i], &t[i + in->Bands], NULL ) )
return( -1 );
if( vips_sum( &t[in->Bands], &t[2 * in->Bands], in->Bands, NULL ) ||
vips_pow_const1( t[2 * in->Bands], out, 0.5, NULL ) )
return( -1 );
return( 0 );
}
static int
vips_smartcrop_attention( VipsSmartcrop *smartcrop,
VipsImage *in, int *left, int *top, int *attention_x, int *attention_y)
{
/* From smartcrop.js.
*/
static double skin_vector[] = {-0.78, -0.57, -0.44};
static double ones[] = {1.0, 1.0, 1.0};
VipsImage **t = (VipsImage **)
vips_object_local_array( VIPS_OBJECT( smartcrop ), 24 );
double hscale;
double vscale;
double sigma;
double max;
int x_pos;
int y_pos;
/* The size we shrink to gives the precision with which we can place
* the crop
*/
hscale = 32.0 / in->Xsize;
vscale = 32.0 / in->Ysize;
sigma = VIPS_MAX( sqrt( pow( smartcrop->width * hscale, 2 ) +
pow( smartcrop->height * vscale, 2 ) ) / 10, 1.0 );
if ( vips_resize( in, &t[17], hscale,
"vscale", vscale,
NULL ) )
return( -1 );
/* Simple edge detect.
*/
if( !(t[21] = vips_image_new_matrixv( 3, 3,
0.0, -1.0, 0.0,
-1.0, 4.0, -1.0,
0.0, -1.0, 0.0 )) )
return( -1 );
/* Convert to XYZ and just use the first three bands.
*/
if( vips_colourspace( t[17], &t[0], VIPS_INTERPRETATION_XYZ, NULL ) ||
vips_extract_band( t[0], &t[1], 0, "n", 3, NULL ) )
return( -1 );
/* Edge detect on Y.
*/
if( vips_extract_band( t[1], &t[2], 1, NULL ) ||
vips_conv( t[2], &t[3], t[21],
"precision", VIPS_PRECISION_INTEGER,
NULL ) ||
vips_linear1( t[3], &t[4], 5.0, 0.0, NULL ) ||
vips_abs( t[4], &t[14], NULL ) )
return( -1 );
/* Look for skin colours. Taken from smartcrop.js.
*/
if(
/* Normalise to magnitude of colour in XYZ.
*/
pythagoras( smartcrop, t[1], &t[5] ) ||
vips_divide( t[1], t[5], &t[6], NULL ) ||
/* Distance from skin point.
*/
vips_linear( t[6], &t[7], ones, skin_vector, 3, NULL ) ||
pythagoras( smartcrop, t[7], &t[8] ) ||
/* Rescale to 100 - 0 score.
*/
vips_linear1( t[8], &t[9], -100.0, 100.0, NULL ) ||
/* Ignore dark areas.
*/
vips_more_const1( t[2], &t[10], 5.0, NULL ) ||
!(t[11] = vips_image_new_from_image1( t[10], 0.0 )) ||
vips_ifthenelse( t[10], t[9], t[11], &t[15], NULL ) )
return( -1 );
/* Look for saturated areas.
*/
if( vips_colourspace( t[1], &t[12],
VIPS_INTERPRETATION_LAB, NULL ) ||
vips_extract_band( t[12], &t[13], 1, NULL ) ||
vips_ifthenelse( t[10], t[13], t[11], &t[16], NULL ) )
return( -1 );
/* Sum, blur and find maxpos.
*
* The amount of blur is related to the size of the crop
* area: how large an area we want to consider for the scoring
* function.
*/
if( vips_sum( &t[14], &t[18], 3, NULL ) ||
vips_gaussblur( t[18], &t[19], sigma, NULL ) ||
vips_max( t[19], &max, "x", &x_pos, "y", &y_pos, NULL ) )
return( -1 );
/* Transform back into image coordinates.
*/
*attention_x = x_pos / hscale;
*attention_y = y_pos / vscale;
/* Centre the crop over the max.
*/
*left = VIPS_CLIP( 0,
*attention_x - smartcrop->width / 2,
in->Xsize - smartcrop->width );
*top = VIPS_CLIP( 0,
*attention_y - smartcrop->height / 2,
in->Ysize - smartcrop->height );
return( 0 );
}
static int
vips_smartcrop_build( VipsObject *object )
{
VipsObjectClass *class = VIPS_OBJECT_GET_CLASS( object );
VipsConversion *conversion = VIPS_CONVERSION( object );
VipsSmartcrop *smartcrop = (VipsSmartcrop *) object;
VipsImage **t = (VipsImage **) vips_object_local_array( object, 2 );
VipsImage *in;
int left;
int top;
int attention_x = 0;
int attention_y = 0;
if( VIPS_OBJECT_CLASS( vips_smartcrop_parent_class )->
build( object ) )
return( -1 );
if( smartcrop->width > smartcrop->in->Xsize ||
smartcrop->height > smartcrop->in->Ysize ||
smartcrop->width <= 0 || smartcrop->height <= 0 ) {
vips_error( class->nickname, "%s", _( "bad extract area" ) );
return( -1 );
}
in = smartcrop->in;
/* If there's an alpha, we have to premultiply before searching for
* content. There could be stuff in transparent areas which we don't
* want to consider.
*/
if( vips_image_hasalpha( in ) ) {
if( vips_premultiply( in, &t[0], NULL ) )
return( -1 );
in = t[0];
}
switch( smartcrop->interesting ) {
case VIPS_INTERESTING_NONE:
case VIPS_INTERESTING_LOW:
left = 0;
top = 0;
break;
case VIPS_INTERESTING_CENTRE:
left = (in->Xsize - smartcrop->width) / 2;
top = (in->Ysize - smartcrop->height) / 2;
break;
case VIPS_INTERESTING_ENTROPY:
if( vips_smartcrop_entropy( smartcrop, in, &left, &top ) )
return( -1 );
break;
case VIPS_INTERESTING_ATTENTION:
if( vips_smartcrop_attention( smartcrop, in, &left, &top, &attention_x, &attention_y ) )
return( -1 );
break;
case VIPS_INTERESTING_HIGH:
left = in->Xsize - smartcrop->width;
top = in->Ysize - smartcrop->height;
break;
case VIPS_INTERESTING_ALL:
left = 0;
top = 0;
smartcrop->width = in->Xsize;
smartcrop->height = in->Ysize;
break;
default:
g_assert_not_reached();
/* Stop a compiler warning.
*/
left = 0;
top = 0;
break;
}
g_object_set(smartcrop,
"attention_x", attention_x,
"attention_y", attention_y,
NULL);
if( vips_extract_area( smartcrop->in, &t[1],
left, top,
smartcrop->width, smartcrop->height, NULL ) ||
vips_image_write( t[1], conversion->out ) )
return( -1 );
return( 0 );
}
static void
vips_smartcrop_class_init( VipsSmartcropClass *class )
{
GObjectClass *gobject_class = G_OBJECT_CLASS( class );
VipsObjectClass *vobject_class = VIPS_OBJECT_CLASS( class );
VIPS_DEBUG_MSG( "vips_smartcrop_class_init\n" );
gobject_class->set_property = vips_object_set_property;
gobject_class->get_property = vips_object_get_property;
vobject_class->nickname = "smartcrop";
vobject_class->description = _( "extract an area from an image" );
vobject_class->build = vips_smartcrop_build;
VIPS_ARG_IMAGE( class, "input", 0,
_( "Input" ),
_( "Input image" ),
VIPS_ARGUMENT_REQUIRED_INPUT,
G_STRUCT_OFFSET( VipsSmartcrop, in ) );
VIPS_ARG_INT( class, "width", 4,
_( "Width" ),
_( "Width of extract area" ),
VIPS_ARGUMENT_REQUIRED_INPUT,
G_STRUCT_OFFSET( VipsSmartcrop, width ),
1, VIPS_MAX_COORD, 1 );
VIPS_ARG_INT( class, "height", 5,
_( "Height" ),
_( "Height of extract area" ),
VIPS_ARGUMENT_REQUIRED_INPUT,
G_STRUCT_OFFSET( VipsSmartcrop, height ),
1, VIPS_MAX_COORD, 1 );
VIPS_ARG_ENUM( class, "interesting", 6,
_( "Interesting" ),
_( "How to measure interestingness" ),
VIPS_ARGUMENT_OPTIONAL_INPUT,
G_STRUCT_OFFSET( VipsSmartcrop, interesting ),
VIPS_TYPE_INTERESTING, VIPS_INTERESTING_ATTENTION );
VIPS_ARG_INT( class, "attention_x", 2,
_( "Attention x" ),
_( "Horizontal position of attention centre" ),
VIPS_ARGUMENT_OPTIONAL_OUTPUT,
G_STRUCT_OFFSET( VipsSmartcrop, attention_x ),
0, VIPS_MAX_COORD, 0 );
VIPS_ARG_INT( class, "attention_y", 3,
_( "Attention y" ),
_( "Vertical position of attention centre" ),
VIPS_ARGUMENT_OPTIONAL_OUTPUT,
G_STRUCT_OFFSET( VipsSmartcrop, attention_y ),
0, VIPS_MAX_COORD, 0 );
}
static void
vips_smartcrop_init( VipsSmartcrop *smartcrop )
{
smartcrop->interesting = VIPS_INTERESTING_ATTENTION;
}
/**
* vips_smartcrop: (method)
* @in: input image
* @out: (out): output image
* @width: width of area to extract
* @height: height of area to extract
* @...: %NULL-terminated list of optional named arguments
*
* Optional arguments:
*
* * @interesting: #VipsInteresting to use to find interesting areas (default: #VIPS_INTERESTING_ATTENTION)
* * @attention_x: %gint, horizontal position of attention centre when using attention based cropping
* * @attention_y: %gint, vertical position of attention centre when using attention based cropping
*
* Crop an image down to a specified width and height by removing boring parts.
*
* Use @interesting to pick the method vips uses to decide which bits of the
* image should be kept.
*
* You can test xoffset / yoffset on @out to find the location of the crop
* within the input image.
*
* See also: vips_extract_area().
*
* Returns: 0 on success, -1 on error.
*/
int
vips_smartcrop( VipsImage *in, VipsImage **out, int width, int height, ... )
{
va_list ap;
int result;
va_start( ap, height );
result = vips_call_split( "smartcrop", ap, in, out, width, height );
va_end( ap );
return( result );
}