Commit 75061040 authored by Antoine Cellerier's avatar Antoine Cellerier

Magnify: validate input chroma, fix for non "YUV 4:2:0" planar chromas, use a...

Magnify: validate input chroma, fix for non "YUV 4:2:0" planar chromas, use a better looking scaling algorithm.
parent 4096b093
...@@ -31,6 +31,8 @@ ...@@ -31,6 +31,8 @@
#include <math.h> #include <math.h>
#include "filter_common.h" #include "filter_common.h"
#include "filter_picture.h"
#include "vlc_image.h" #include "vlc_image.h"
#include "vlc_input.h" #include "vlc_input.h"
#include "vlc_playlist.h" #include "vlc_playlist.h"
...@@ -93,6 +95,16 @@ static int Create( vlc_object_t *p_this ) ...@@ -93,6 +95,16 @@ static int Create( vlc_object_t *p_this )
{ {
vout_thread_t *p_vout = (vout_thread_t *)p_this; vout_thread_t *p_vout = (vout_thread_t *)p_this;
switch( p_vout->fmt_in.i_chroma )
{
CASE_PLANAR_YUV
case VLC_FOURCC('G','R','E','Y'):
break;
default:
msg_Err( p_vout, "Unsupported chroma" );
return VLC_EGENERIC;
}
/* Allocate structure */ /* Allocate structure */
p_vout->p_sys = malloc( sizeof( vout_sys_t ) ); p_vout->p_sys = malloc( sizeof( vout_sys_t ) );
if( p_vout->p_sys == NULL ) if( p_vout->p_sys == NULL )
...@@ -149,8 +161,8 @@ static int Init( vout_thread_t *p_vout ) ...@@ -149,8 +161,8 @@ static int Init( vout_thread_t *p_vout )
#define VIS_ZOOM 4 #define VIS_ZOOM 4
p_vout->p_sys->i_x = 0; p_vout->p_sys->i_x = 0;
p_vout->p_sys->i_y = 0; p_vout->p_sys->i_y = 0;
#define ZOOM_FACTOR 100 #define ZOOM_FACTOR 8
p_vout->p_sys->i_zoom = 200; p_vout->p_sys->i_zoom = 2*ZOOM_FACTOR;
p_vout->p_sys->b_visible = VLC_TRUE; p_vout->p_sys->b_visible = VLC_TRUE;
var_AddCallback( p_vout->p_sys->p_vout, "mouse-x", MouseEvent, p_vout ); var_AddCallback( p_vout->p_sys->p_vout, "mouse-x", MouseEvent, p_vout );
...@@ -220,6 +232,7 @@ static void Render( vout_thread_t *p_vout, picture_t *p_pic ) ...@@ -220,6 +232,7 @@ static void Render( vout_thread_t *p_vout, picture_t *p_pic )
video_format_t fmt_out; video_format_t fmt_out;
picture_t *p_converted; picture_t *p_converted;
plane_t *p_oyp=NULL; plane_t *p_oyp=NULL;
int i_plane;
memset( &fmt_out, 0, sizeof(video_format_t) ); memset( &fmt_out, 0, sizeof(video_format_t) );
/* This is a new frame. Get a structure from the video_output. */ /* This is a new frame. Get a structure from the video_output. */
...@@ -240,35 +253,114 @@ static void Render( vout_thread_t *p_vout, picture_t *p_pic ) ...@@ -240,35 +253,114 @@ static void Render( vout_thread_t *p_vout, picture_t *p_pic )
/* background magnified image */ /* background magnified image */
if( o_zoom != ZOOM_FACTOR ) if( o_zoom != ZOOM_FACTOR )
{ {
#define magnify( plane ) \ for( i_plane = 0; i_plane < p_pic->i_planes; i_plane++ )
o_yp = o_y*p_outpic->p[plane].i_lines/p_outpic->p[Y_PLANE].i_lines; \ {
o_xp = o_x*p_outpic->p[plane].i_pitch/p_outpic->p[Y_PLANE].i_pitch; \ o_yp = o_y*p_outpic->p[i_plane].i_lines
for( y=0; y<p_outpic->p[plane].i_visible_lines; y++ ) \ /p_outpic->p[Y_PLANE].i_lines;
{ \ o_xp = o_x*p_outpic->p[i_plane].i_pitch
for( x=0; x<p_outpic->p[plane].i_visible_pitch; x++ ) \ /p_outpic->p[Y_PLANE].i_pitch;
{ \ int i_pitch = p_outpic->p[i_plane].i_pitch;
p_outpic->p[plane].p_pixels[y*p_outpic->p[plane].i_pitch+x] = \ #if 0
p_pic->p[plane].p_pixels[ \ int o_zoom2 = o_zoom*o_zoom;
( o_yp + y*ZOOM_FACTOR/o_zoom )*p_outpic->p[plane].i_pitch \ int o_zoom3 = o_zoom*o_zoom*o_zoom;
+ o_xp + x*ZOOM_FACTOR/o_zoom \ int o_zoom4 = o_zoom*o_zoom*o_zoom*o_zoom;
]; \ int o_zoom6 = o_zoom*o_zoom*o_zoom * o_zoom*o_zoom*o_zoom;
} \ #endif
} for( y=0; y<p_outpic->p[i_plane].i_visible_lines; y++ )
magnify( Y_PLANE ); {
magnify( U_PLANE ); for( x=0; x<p_outpic->p[i_plane].i_visible_pitch; x++ )
magnify( V_PLANE ); {
#undef magnify #if 0
/* Nearest neighbor */
int nx = o_xp + x*ZOOM_FACTOR/o_zoom;
int ny = o_yp + y*ZOOM_FACTOR/o_zoom;
p_outpic->p[i_plane].p_pixels[y*i_pitch+x] =
p_pic->p[i_plane].p_pixels[ny*i_pitch+nx];
#elif 1
/* Bi-linear */
int nx_real = o_xp*o_zoom + x*ZOOM_FACTOR;
int ny_real = o_yp*o_zoom + y*ZOOM_FACTOR;
int nx = nx_real/o_zoom;
int ny = ny_real/o_zoom;
int wtl = ((nx+1)*o_zoom-nx_real)*((ny+1)*o_zoom-ny_real);
int wtr = (nx_real-nx*o_zoom)*((ny+1)*o_zoom-ny_real);
int wbl = ((nx+1)*o_zoom-nx_real)*(ny_real-ny*o_zoom);
int wbr = (nx_real-nx*o_zoom)*(ny_real-ny*o_zoom);
p_outpic->p[i_plane].p_pixels[y*i_pitch+x] =
( wtl*p_pic->p[i_plane].p_pixels[ny*i_pitch+nx]
+ wtr*p_pic->p[i_plane].p_pixels[ny*i_pitch+(nx+1)]
+ wbl*p_pic->p[i_plane].p_pixels[(ny+1)*i_pitch+nx]
+ wbr*p_pic->p[i_plane].p_pixels[(ny+1)*i_pitch+(nx+1)]
) / (o_zoom*o_zoom);
#else
/* Bi-cubic */
/* FIXME: doesn't work */
/* \Sigma_{i=0..3} \Sigma_{j=0..3} \alpha_{i,j} x^i y^j
* 16 \alpha_{i,j} so we need to use the 16 nearest pixels
*
* In fact we should be able to write it as:
* \Pi_{i=1..16} [ \Pi_{j!=i} ( x - x_j ) * ( y - y_j ) ]
* * z_i / [ \Pi_{j!=i} ( x_i - x_j ) * ( y_i - y_j ) ]
* We also have to make sure that we don't use any
* x_j == x_i or y_j == y_i in the \Pi on j (else we get
* a 0/0 which kind of sucks) .
*/
uint8_t *p = p_pic->p[i_plane].p_pixels;
int nx_real = o_xp*o_zoom + x*ZOOM_FACTOR;
int ny_real = o_yp*o_zoom + y*ZOOM_FACTOR;
int nx = nx_real/o_zoom;
int ny = ny_real/o_zoom;
int xi, yi, xj, yj;
int my = __MAX( ny-1, 0 );
int mx = __MAX( nx-1, 0 );
//p_outpic->p[i_plane].i_visible_lines - 4
//p_outpic->p[i_plane].i_visible_pitch - 4
int v = 1;
for( yi = my; yi <= my+3; yi++ )
{
int numy = 1;
int deny = 1;
/*
for( yj = my; yj <= my+3; yj++ )
{
if( yj != yi )
{
numy *= (ny_real-yj*o_zoom);
deny *= (yi - yj);
}
}
numy /= o_zoom2;*/
for( xi = mx; xi <= mx+3; xi++ )
{
int num = numy;
int den = deny;
for( xj = mx; xj <= mx+3; xj++ )
{
if( xj != xi )
{
num *= (nx_real-xj*o_zoom);
den *= (xi - xj);
}
}
v = ( v * (p[yi*i_pitch+xi] * num) ) / ( den * o_zoom3 );
}
}
p_outpic->p[i_plane].p_pixels[y*i_pitch+x] = v;
#endif
}
}
}
} }
else else
{ {
#define copy( plane ) \ for( i_plane = 0; i_plane < p_pic->i_planes; i_plane++ )
p_vout->p_libvlc-> \ {
pf_memcpy( p_outpic->p[plane].p_pixels, p_pic->p[plane].p_pixels, \ p_vout->p_libvlc->
p_outpic->p[plane].i_lines * p_outpic->p[plane].i_pitch ); pf_memcpy( p_outpic->p[i_plane].p_pixels, p_pic->p[i_plane].p_pixels,
copy( Y_PLANE ); p_outpic->p[i_plane].i_lines * p_outpic->p[i_plane].i_pitch );
copy( U_PLANE ); }
copy( V_PLANE );
#undef copy
} }
if( p_vout->p_sys->b_visible ) if( p_vout->p_sys->b_visible )
...@@ -279,18 +371,16 @@ static void Render( vout_thread_t *p_vout, picture_t *p_pic ) ...@@ -279,18 +371,16 @@ static void Render( vout_thread_t *p_vout, picture_t *p_pic )
fmt_out.i_height = p_vout->render.i_height/VIS_ZOOM; fmt_out.i_height = p_vout->render.i_height/VIS_ZOOM;
p_converted = image_Convert( p_vout->p_sys->p_image, p_pic, p_converted = image_Convert( p_vout->p_sys->p_image, p_pic,
&(p_pic->format), &fmt_out ); &(p_pic->format), &fmt_out );
#define copyimage( plane ) \ for( i_plane = 0; i_plane < p_pic->i_planes; i_plane++ )
for( y=0; y<p_converted->p[plane].i_visible_lines; y++) \ {
{ \ for( y=0; y<p_converted->p[i_plane].i_visible_lines; y++)
p_vout->p_libvlc->pf_memcpy( \ {
p_outpic->p[plane].p_pixels+y*p_outpic->p[plane].i_pitch, \ p_vout->p_libvlc->pf_memcpy(
p_converted->p[plane].p_pixels+y*p_converted->p[plane].i_pitch, \ p_outpic->p[i_plane].p_pixels+y*p_outpic->p[i_plane].i_pitch,
p_converted->p[plane].i_visible_pitch ); \ p_converted->p[i_plane].p_pixels+y*p_converted->p[i_plane].i_pitch,
p_converted->p[i_plane].i_visible_pitch );
}
} }
copyimage( Y_PLANE );
copyimage( U_PLANE );
copyimage( V_PLANE );
#undef copyimage
p_converted->pf_release( p_converted ); p_converted->pf_release( p_converted );
/* white rectangle on visualization */ /* white rectangle on visualization */
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment