3dpcp/.svn/pristine/29/298c79a7583e5f6fe0e26fd4a112735e64c30189.svn-base
2012-09-16 14:33:11 +02:00

703 lines
28 KiB
Text

/*
This is a Optical-Character-Recognition program
Copyright (C) 2000-2009 Joerg Schulenburg
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU 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 General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
see README for EMAIL-address
*/
#include <stdlib.h>
#include <stdio.h>
#include "pgm2asc.h"
#include "gocr.h"
#include "progress.h"
/* measure mean thickness as an criteria for big chars */
int mean_thickness( struct box *box2 ){
int mt=0, i, y, dx=box2->x1-box2->x0+1, dy;
for (y=box2->y0+1; y<box2->y1; y++) {
i=loop(box2->p,box2->x0+0,y,dx,JOB->cfg.cs,0,RI);
i=loop(box2->p,box2->x0+i,y,dx,JOB->cfg.cs,1,RI);
mt+=i;
}
dy = box2->y1 - box2->y0 - 1;
if (dy) mt=(mt+dy/2)/dy;
return mt;
}
/* ---- remove dust ---------------------------------
What is dust? I think, this is a very small pixel cluster without
neighbours. Of course not all dust clusters can be detected correct.
This feature should be possible to switch off via option.
-> may be, all clusters should be stored here?
speed is very slow, I know, but I am happy that it is working well
*/
int remove_dust( job_t *job ){
/* new dust removing */
/* FIXME jb:remove pp */
pix *pp = &job->src.p;
int i1,i,j,x,y,x0,x1,y0,y1,nC,sX,sY,sP, cs,vvv=job->cfg.verbose;
struct box *box2;
#define HISTSIZE 220 /* histogramm size */
int histo[HISTSIZE];
cs=job->cfg.cs; sP=sX=sY=nC=0;
/*
* count number of black pixels within a box and store it in .dots
* later .dots is re-used for number of objects belonging to the character
* should be done in the flood-fill algorithm
* volume of white pixels is estimated to big here (left/right rot)
* ToDo: mean thickness of char lines?
* or interval nesting (minP..maxP) to remove outriders
*/
j=0;
for (i1=0;i1<HISTSIZE;i1++) histo[i1]=0;
/* mean value over every black object which is big enough */
for_each_data(&(job->res.boxlist)) {
box2 = (struct box *)list_get_current(&(job->res.boxlist));
if (!box2->num_frames) continue;
if (box2->frame_vol[0]<0) continue; /* don't count inner holes */
j = abs(box2->frame_vol[0]);
if ((box2->y1-box2->y0+1)>3) {
nC++; /* only count potential chars v0.42 */
sX+=box2->x1 - box2->x0 + 1;
sY+=box2->y1 - box2->y0 + 1;
sP+=j;
}
if (j<HISTSIZE) histo[j]++;
} end_for_each(&(job->res.boxlist));
if (job->cfg.dust_size < 0 && nC > 0) { /* auto detection */
/* this formula is empirically, high resolution scans have bigger dust */
/* maximum allowed dustsize (min=4*7 ca. 32)
* does not work for background pattern!
*/
job->cfg.dust_size = ( ( sX/nC ) * ( sY/nC ) + 16) / 32;
if (vvv) fprintf(stderr, "# remove.c remove_dust(): ");
if (vvv) fprintf(stderr, "\n# dust size detection, vol num"
" #obj=%d maxDust=%d mpixel= %3d mxy= %2d %2d",
nC, job->cfg.dust_size, sP/nC, sX/nC, sY/nC);
/* we assume that for random dust applies histo[i+1]<histo[i] */
for (i=1;i+3<HISTSIZE;i++){
if (vvv) fprintf(stderr,"\n# dust size histogram %3d %5d",i,histo[i]);
if (histo[i]>=nC) continue; /* v0.42 lot of pixels -> bg pattern < 3 */
if (i>=job->cfg.dust_size) break; /* maximum = mean size / 32 */
if (histo[i/*+1*/]==0) break; /* bad statistic */
if ((histo[i+2]+histo[i+3])
>=(histo[i] +histo[i+1])) break; /* no noise, but to late? */
if ( histo[i-1] > 1024*histo[i] &&
2*histo[i+1] >=histo[i]) break; /* bg pattern */
}
if (vvv) fprintf(stderr," break");
if (vvv) for (i1=0,j=i+1;j<HISTSIZE;j++) {
/* compressed, output only if something is changing */
if (j==HISTSIZE-1 || histo[j]!=histo[j-1] || histo[j]!=histo[j+1]) {
fprintf(stderr,"\n# dust size histogram %3d %5d",j,histo[j]);
if (++i1>20) break; /* dont do excessive output */
}
}
job->cfg.dust_size=i-1;
/* what is the statistic of random dust?
* if we have p pixels on a x*y image we should have
* (p/(x*y))^1 * (x*y) = p singlets
* (p/(x*y))^2 * (x*y) = p^2/(x*y) doublets and
* (p/(x*y))^3 * (x*y) = p^3/(x*y)^2 triplets
*/
if (vvv) fprintf(stderr,"\n# auto dust size = %d nC= %3d .. %3d"
" avD= %2d %2d .. %2d %2d\n",
job->cfg.dust_size, nC, job->res.numC,
(job->res.sumX+job->res.numC/2)/job->res.numC,
(job->res.sumY+job->res.numC/2)/job->res.numC, sX/nC, sY/nC);
}
if (job->cfg.dust_size)
{ i=0;
if(vvv){
fprintf(stderr,"# remove dust of size %2d",job->cfg.dust_size);
/* Warning: better use (1/(x*y))^2 as 1/((x*y)^2),
* because (x*y)^2 may overflow */
fprintf(stderr," histo=%d,%d(?=%d),%d(?=%d),...\n# ...",
histo[1],histo[2],histo[1]*histo[1]/(pp->x*pp->y),
histo[3], histo[1]*histo[1]/(pp->x*pp->y)
*histo[1]/(pp->x*pp->y));
}
i = 0;
for_each_data(&(job->res.boxlist)) {
box2 = (struct box *)list_get_current(&(job->res.boxlist));
x0=box2->x0;x1=box2->x1;y0=box2->y0;y1=box2->y1; /* box */
j=abs(box2->frame_vol[0]);
if(j<=job->cfg.dust_size) /* remove this tiny object */
{ /* here we should distinguish dust and i-dots,
* may be we should sort out dots to a seperate dot list and
* after line detection decide, which is dust and which not
* dust should be removed to make recognition easier (ToDo)
*/
#if 0
if(get_bw((3*x0+x1)/4,(x0+3*x1)/4,y1+y1-y0+1,y1+8*(y1-y0+1),pp,cs,1))
continue; /* this idea was to simple, see kscan003.jpg sample */
#endif
/* remove from average */
job->res.numC--;
job->res.sumX-=x1-x0+1;
job->res.sumY-=y1-y0+1;
/* remove pixels (should only be done with dust) */
for(x=x0;x<=x1;x++)
for(y=y0;y<=y1;y++){ put(pp,x,y,0,255&~7); }
/* remove from list */
list_del(&(job->res.boxlist),box2);
/* free memory */
free_box(box2);
i++; /* count as dust particle */
continue;
}
} end_for_each(&(job->res.boxlist));
if(vvv)fprintf(stderr," %3d cluster removed, nC= %3d\n",i,job->res.numC);
}
/* reset dots to 0 and remove white pixels (new) */
i=0;
for_each_data(&(job->res.boxlist)) {
box2 = ((struct box *)list_get_current(&(job->res.boxlist)));
if (box2->frame_vol[0]<0) continue; /* for black areas only */
x0=box2->x0;x1=box2->x1;y0=box2->y0;y1=box2->y1; /* box */
if (x1-x0>16 && y1-y0>30) /* only on large enough chars */
for(x=x0+1;x<=x1-1;x++)
for(y=y0+1;y<=y1-1;y++){
if( pixel_atp(pp,x ,y )>=cs
&& pixel_atp(pp,x-1,y ) <cs
&& pixel_atp(pp,x+1,y ) <cs
&& pixel_atp(pp,x ,y-1) <cs
&& pixel_atp(pp,x ,y+1) <cs ) /* remove it */
{
put(pp,x,y,0,0); i++; /* (x and 0) or 0 */
}
}
} end_for_each(&(job->res.boxlist));
if (vvv) fprintf(stderr,"# ... %3d white pixels removed, cs=%d nC= %3d\n",
i,cs,job->res.numC);
return 0;
}
/* ---- smooth big chars ---------------------------------
* Big chars often do not have smooth borders, which let fail
* the engine. Here we smooth the borders of big chars (>7x16).
* Smoothing is important for b/w scans, where we often have
* comb like pattern on a vertikal border. I also received
* samples with lot of white pixels (sample: 04/02/25).
* ToDo: obsolete if vector code is complete
*/
int smooth_borders( job_t *job ){
pix *pp = &job->src.p;
int ii=0,x,y,x0,x1,y0,y1,dx,dy,cs,i0,i1,i2,i3,i4,n1,n2,
cn[8],cm,vvv=job->cfg.verbose; /* dust found */
struct box *box2;
cs=job->cfg.cs; n1=n2=0;
if(vvv){ fprintf(stderr,"# smooth big chars 7x16 cs=%d",cs); }
/* filter for each big box */
for_each_data(&(job->res.boxlist)) { n2++; /* count boxes */
box2 = (struct box *)list_get_current(&(job->res.boxlist));
/* do not touch small characters! but how we define small characters? */
if (box2->x1-box2->x0+1<7 || box2->y1-box2->y0+1<16 ) continue;
if (box2->c==PICTURE) continue;
if (mean_thickness(box2)<3) continue;
n1++; /* count boxes matching big-char criteria */
x0=box2->x0; y0=box2->y0;
x1=box2->x1; y1=box2->y1;
dx=x1-x0+1; dy=y1-y0-1;
/* out_x(box2);
* dont change to much! only change if absolutely sure!
* ....... 1 2 3
* ex: .?##### 0 * 4
* ....... 7 6 5
* we should also avoid removing lines by sytematic remove
* from left end to the right, so we concern also about distance>1
*/
for(x=box2->x0;x<=box2->x1;x++)
for(y=box2->y0;y<=box2->y1;y++){ /* filter out high frequencies */
/* this is a very primitive solution, only for learning */
cn[0]=getpixel(pp,x-1,y);
cn[4]=getpixel(pp,x+1,y); /* horizontal */
cn[2]=getpixel(pp,x,y-1);
cn[6]=getpixel(pp,x,y+1); /* vertical */
cn[1]=getpixel(pp,x-1,y-1);
cn[3]=getpixel(pp,x+1,y-1); /* diagonal */
cn[7]=getpixel(pp,x-1,y+1);
cn[5]=getpixel(pp,x+1,y+1);
cm=getpixel(pp,x,y);
/* check for 5 other and 3 same surrounding pixels */
for (i0=0;i0<8;i0++)
if ((cn[i0 ]<cs)==(cm<cs)
&& (cn[(i0+7) & 7]<cs)!=(cm<cs)) break; /* first same */
for (i1=0;i1<8;i1++)
if ((cn[(i0+i1) & 7]<cs)!=(cm<cs)) break; /* num same */
for (i2=0;i2<8;i2++)
if ((cn[(i0+i1+i2) & 7]<cs)==(cm<cs)) break; /* num other */
cn[0]=getpixel(pp,x-2,y);
cn[4]=getpixel(pp,x+2,y); /* horizontal */
cn[2]=getpixel(pp,x,y-2);
cn[6]=getpixel(pp,x,y+2); /* vertical */
cn[1]=getpixel(pp,x-2,y-2);
cn[3]=getpixel(pp,x+2,y-2); /* diagonal */
cn[7]=getpixel(pp,x-2,y+2);
cn[5]=getpixel(pp,x+2,y+2);
/* check for 5 other and 3 same surrounding pixels */
for (i0=0;i0<8;i0++)
if ((cn[i0 ]<cs)==(cm<cs)
&& (cn[(i0+7) & 7]<cs)!=(cm<cs)) break; /* first same */
for (i3=0;i3<8;i3++)
if ((cn[(i0+i3) & 7]<cs)!=(cm<cs)) break; /* num same */
for (i4=0;i4<8;i4++)
if ((cn[(i0+i3+i4) & 7]<cs)==(cm<cs)) break; /* num other */
if (i1<=3 && i2>=5 && i3>=3 && i4>=3) { /* change only on borders */
ii++; /* white : black */
put(pp,x,y,7,((cm<cs)?(cs|32):cs/2)&~7);
#if 0
printf(" x y i0 i1 i2 i3 i4 cm new cs %3d %3d"
" %3d %3d %3d %3d %3d %3d %3d %3d\n",
x-box2->x0,y-box2->y0,i0,i1,i2,i3,i3,cm,getpixel(pp,x,y),cs);
#endif
}
}
#if 0 /* debugging */
out_x(box2);
#endif
} end_for_each(&(job->res.boxlist));
if(vvv)fprintf(stderr," ... %3d changes in %d of %d\n",ii,n1,n2);
return 0;
}
/* test if a corner of box1 is within box2 */
int box_nested( struct box *box1, struct box *box2){
/* box1 in box2, +1..-1 frame for pixel-patterns */
if ( ( ( box1->x0>=box2->x0-1 && box1->x0<=box2->x1+1 )
|| ( box1->x1>=box2->x0-1 && box1->x1<=box2->x1+1 ) )
&& ( ( box1->y0>=box2->y0-1 && box1->y0<=box2->y1+1 )
|| ( box1->y1>=box2->y0-1 && box1->y1<=box2->y1+1 ) ) )
return 1;
return 0;
}
/* test if box1 is within box2 */
int box_covered( struct box *box1, struct box *box2){
/* box1 in box2, +1..-1 frame for pixel-patterns */
if ( ( box1->x0>=box2->x0-1 && box1->x1<=box2->x1+1 )
&& ( box1->y0>=box2->y0-1 && box1->y1<=box2->y1+1 ) )
return 1;
return 0;
}
/* ---- remove pictures ------------------------------------------
* may be, not deleting or moving to another list is much better!
* should be renamed to remove_pictures and border boxes
*/
int remove_pictures( job_t *job){
struct box *box4,*box2;
int j=0, j2=0, num_del=0;
if (job->cfg.verbose)
fprintf(stderr, "# "__FILE__" L%d: remove pictures\n# ...",
__LINE__);
/* ToDo: output a list for picture handle scripts */
j=0; j2=0;
if(job->cfg.verbose)
for_each_data(&(job->res.boxlist)) {
box4 = (struct box *)list_get_current(&(job->res.boxlist));
if (box4->c==PICTURE) j++; else j2++;
} end_for_each(&(job->res.boxlist));
if (job->cfg.verbose)
fprintf(stderr," status: pictures= %d other= %d nC= %d\n# ...",
j, j2, job->res.numC);
/* remove table frames */
if (job->res.numC > 8)
for_each_data(&(job->res.boxlist)) {
box2 = (struct box *)list_get_current(&(job->res.boxlist));
if (box2->c==PICTURE
&& box2->num_ac==0 /* dont remove barcodes */
&& box2->x1-box2->x0+1>box2->p->x/2 /* big table? */
&& box2->y1-box2->y0+1>box2->p->y/2 ){ j=0;
/* count boxes nested with the picture */
for_each_data(&(job->res.boxlist)) {
box4 = (struct box *)list_get_current(&(job->res.boxlist));
if( box4 != box2 ) /* not count itself */
if (box_nested(box4,box2)) j++; /* box4 in box2 */
} end_for_each(&(job->res.boxlist));
if( j>8 ){ /* remove box if more than 8 chars are within box */
list_del(&(job->res.boxlist), box2); /* does not work proper ?! */
free_box(box2); num_del++;
}
}
} end_for_each(&(job->res.boxlist));
if (job->cfg.verbose)
fprintf(stderr, " deleted= %d pictures (table frames)\n# ...",
num_del);
num_del=0;
/* remove dark-border-boxes (typical for hard copy of book site,
* or spam random border) */
if (job->res.numC > 1) /* dont remove the only char */
for_each_data(&(job->res.boxlist)) {
box2 = (struct box *)list_get_current(&(job->res.boxlist));
if (box2->c!=PICTURE) continue; // ToDo: PICTUREs set already?
if ( box2->x1-box2->x0+1 > box2->p->x/2
&& box2->y1-box2->y0+1 > box2->p->y/2 ) continue;
j=0;
if (box2->x0==0) j++;
if (box2->y0==0) j++; /* on border? */
if (box2->x1==box2->p->x-1) j++;
if (box2->y1==box2->p->y-1) j++;
if (j>2){ /* ToDo: check corner pixel */
int cs=job->cfg.cs;
j=0;
if (getpixel(box2->p,box2->x0,box2->y0)<cs) j++;
if (getpixel(box2->p,box2->x1,box2->y0)<cs) j++;
if (getpixel(box2->p,box2->x0,box2->y1)<cs) j++;
if (getpixel(box2->p,box2->x1,box2->y1)<cs) j++;
if (j>2) {
list_del(&(job->res.boxlist), box2);
free_box(box2); num_del++;
}
}
} end_for_each(&(job->res.boxlist));
if (job->cfg.verbose)
fprintf(stderr, " deleted= %d pictures (on border)\n# ...",
num_del);
num_del=0;
j=0; j2=0;
if(job->cfg.verbose)
for_each_data(&(job->res.boxlist)) {
box4 = (struct box *)list_get_current(&(job->res.boxlist));
if( box4->c==PICTURE ) j++; else j2++;
} end_for_each(&(job->res.boxlist));
if (job->cfg.verbose)
fprintf(stderr," status: pictures= %d other= %d nC= %d\n# ...",
j, j2, job->res.numC);
for(j=1;j;){ j=0; /* this is only because list_del does not work */
/* can be slow on gray images */
for_each_data(&(job->res.boxlist)) {
box2 = (struct box *)list_get_current(&(job->res.boxlist));
if( box2->c==PICTURE && box2->num_ac==0)
for(j=1;j;){ /* let it grow to max before leave */
j=0; box4=NULL;
/* find boxes nested with the picture and remove */
/* its for pictures build by compounds */
for_each_data(&(job->res.boxlist)) {
box4 = (struct box *)list_get_current(&(job->res.boxlist));
if( box4!=box2 /* not destroy self */
&& (box4->num_ac==0) /* dont remove barcodes etc. */
&& (/* box4->c==UNKNOWN || */
box4->c==PICTURE) ) /* dont remove valid chars */
if(
/* box4 in box2, +1..-1 frame for pixel-patterns */
box_nested(box4,box2)
/* or box2 in box4 */
|| box_nested(box2,box4) /* same? */
)
if ( box4->x1-box4->x0+1>2*job->res.avX
|| box4->x1-box4->x0+1<job->res.avX/2
|| box4->y1-box4->y0+1>2*job->res.avY
|| box4->y1-box4->y0+1<job->res.avY/2
|| box_covered(box4,box2) ) /* box4 completely within box2 */
/* dont remove chars! see rotate45.fig */
{
/* do not remove boxes in inner loop (bug?) ToDo: check why! */
/* instead we leave inner loop and mark box4 as valid */
if( box4->x0<box2->x0 ) box2->x0=box4->x0;
if( box4->x1>box2->x1 ) box2->x1=box4->x1;
if( box4->y0<box2->y0 ) box2->y0=box4->y0;
if( box4->y1>box2->y1 ) box2->y1=box4->y1;
j=1; /* mark box4 as valid */
break; /* and leave inner loop */
}
} end_for_each(&(job->res.boxlist));
if (j!=0 && box4!=NULL) { /* check for valid box4 */
/* ToDo: melt */
list_del(&(job->res.boxlist), box4); /* does not work proper ?! */
free_box(box4); /* break; ToDo: necessary to leave after del??? */
num_del++;
}
}
} end_for_each(&(job->res.boxlist));
}
if (job->cfg.verbose)
fprintf(stderr, " deleted= %d nested pictures\n# ...", num_del);
/* output a list for picture handle scripts */
j=0; j2=0;
if(job->cfg.verbose)
for_each_data(&(job->res.boxlist)) {
box4 = (struct box *)list_get_current(&(job->res.boxlist));
if( box4->c==PICTURE ) {
fprintf(stderr," found picture at %4d %4d size %4d %4d\n# ...",
box4->x0, box4->y0, box4->x1-box4->x0+1, box4->y1-box4->y0+1 );
j++;
} else j2++;
} end_for_each(&(job->res.boxlist));
if (job->cfg.verbose)
fprintf(stderr," status: pictures= %d other= %d nC= %d\n",
j, j2, job->res.numC);
return 0;
}
/* ---- remove melted serifs --------------------------------- v0.2.5
>>v<<
##########.######## <-y0
################### like X VW etc.
...###.......###... <-y
...###......###....
j1 j2 j3
- can generate new boxes if two characters were glued
*/
int remove_melted_serifs( pix *pp ){
int x,y,j1,j2,j3,j4,i2,i3,i,ii,ni,cs,x0,x1,xa,xb,y0,y1,vvv=JOB->cfg.verbose;
struct box *box2, *box3;
progress_counter_t *pc = NULL;
cs=JOB->cfg.cs; i=0; ii=0; ni=0;
for_each_data(&(JOB->res.boxlist)) {
ni++;
} end_for_each(&(JOB->res.boxlist));
pc = open_progress(ni,"remove_melted_serifs");
ni = 0;
if(vvv){ fprintf(stderr,"# searching melted serifs ..."); }
for_each_data(&(JOB->res.boxlist)) {
box2 = (struct box *)list_get_current(&(JOB->res.boxlist));
if (box2->c != UNKNOWN) continue; /* dont try on pictures */
x0=box2->x0; x1=box2->x1;
y0=box2->y0; y1=box2->y1; /* box */
/* upper serifs */
for(j1=x0;j1+4<x1;){
j1+=loop(pp,j1,y0 ,x1-x0,cs,0,RI);
x =loop(pp,j1,y0 ,x1-x0,cs,1,RI); if(j1+x>x1+1) break;
y =loop(pp,j1,y0+1,x1-x0,cs,1,RI); if(y>x) x=y; if(j1+x>x1+1) break;
/* measure mean thickness of serif pos: (j1,y0)-(j1+x,y0) */
for(j2=j3=j4=0,i2=j1;i2<j1+x;i2++){
/* 2009-07: bug, j1 used instead of i2 */
i3 =loop(pp,i2,y0 ,y1-y0,cs,0,DO); if(8*i3>y1-y0) break;
i3+=loop(pp,i2,y0+i3,y1-y0,cs,1,DO); if(8*i3>y1-y0) continue;
if(8*i3<y1-y0){ j2+=i3; j3++; } /* sum vert. thickness */
} if(j3==0){ j1+=x; continue; } /* no serif, skip this object */
y = y0+(j2+j3-1)/j3+(y1-y0+1)/32; /* y0 + mean thickness + dy/32 + 1 */
if (vvv&1)
fprintf(stderr, "\n# upper serif x0,y0,j1-x0+x,y-y0 %4d %4d %2d+%2d %2d",
x0,y0,j1-x0,x,y-y0);
/* check if really melted serifs */
if (loop(pp,j1,y,x1-x0,cs,0,RI)<1) { j1+=x; continue; }
if(num_cross(j1 ,j1+x,y,y,pp,cs) < 2 ){ j1+=x;continue; }
if (vvv&1)
fprintf(stderr, " ok1");
j2 = j1 + loop(pp,j1,y,x1-x0,cs,0,RI);
j2 = j2 + loop(pp,j2,y,x1-x0,cs,1,RI);
i3 = loop(pp,j2,y,x1-x0,cs,0,RI); if(i3<2){j1+=x;continue;}
j2 += i3/2;
j3 = j2 + loop(pp,j2,y ,x1-j2,cs,0,RI);
i3 = j2 + loop(pp,j2,y+1,x1-j2,cs,0,RI); if(i3>j3)j3=i3;
j3 = j3 + loop(pp,j3,y ,x1-j3,cs,1,RI);
i3 = loop(pp,j3,y ,x1-j3,cs,0,RI);
if(i3<2 || j3>=j1+x){j1+=x;continue;}
j3 += i3/2;
if(x>5)
{
i++; /* snip! */
for(y=0;y<(y1-y0+1+4)/8;y++)put(pp,j2,y0+y,255,128+64); /* clear highest bit */
if(vvv&4){
fprintf(stderr,"\n");
out_x(box2);
fprintf(stderr,"# melted serifs corrected on %d %d j1=%d j3=%d",
j2-x0, y, j1-x0, j3-x0);
// ToDo: vector cut with line from xa,ya to xb,yb
// two frames of double melted MN become one frame if cut one
// of the melted serifs (new function cut_frames_at_line())
}
for(xb=0,xa=0;xa<(x1-x0+4)/8;xa++){ /* detect vertical gap */
i3=y1;
if(box2->m3>y0 && 2*y1>box2->m3+box2->m4) i3=box2->m3; /* some IJ */
if( loop(pp,j2-xa,i3,i3-y0,cs,0,UP) > (y1-y0+1)/2
&& loop(pp,j2,(y0+y1)/2,xa+1,cs,0,LE) >=xa ){ xb=-xa; break; }
if( loop(pp,j2+xa,i3,i3-y0,cs,0,UP) > (y1-y0+1)/2
&& loop(pp,j2,(y0+y1)/2,xa+1,cs,0,RI) >=xa ){ xb= xa; break; }
}
if( get_bw(j2 ,j2 ,y0,(y0+y1)/2,pp,cs,1) == 0
&& get_bw(j2+xb,j2+xb,(y0+y1)/2,i3,pp,cs,1) == 0 )
{ /* divide */
box3=malloc_box(box2);
box3->x1=j2-1;
box2->x0=j2+1; x1=box2->x1;
cut_box(box2); /* cut vectors outside the box, see box.c */
cut_box(box3);
box3->num=JOB->res.numC;
list_ins(&(JOB->res.boxlist),box2,box3); JOB->res.numC++; ii++; /* insert box3 before box2 */
if(vvv&4) fprintf(stderr," => splitted");
j1=x0=box2->x0; x=0; /* hopefully ok, UVW */
}
}
j1+=x;
}
/* same on lower serifs -- change this later to better function
// #### ###
// #### v ### # <-y
// #################### <-y1
// j1 j2 j3
*/
for(j1=x0;j1<x1;){
j1+=loop(pp,j1,y1 ,x1-x0,cs,0,RI);
x =loop(pp,j1,y1 ,x1-x0,cs,1,RI); if(j1+x>x1+1) break;
y =loop(pp,j1,y1-1,x1-x0,cs,1,RI); if(y>x) x=y; if(j1+x>x1+1) break;
/* measure mean thickness of serif */
for(j2=j3=j4=0,i2=j1;i2<j1+x;i2++){
/* 2009-07: bug, j1 used instead of i2 */
i3 =loop(pp,i2,y1 ,y1-y0,cs,0,UP); if(8*i3>y1-y0) break;
i3+=loop(pp,i2,y1-i3,y1-y0,cs,1,UP); if(8*i3>y1-y0) continue;
if(8*i3<y1-y0){ j2+=i3; j3++; }
} if(j3==0){ j1+=x; continue; }
y = y1-(j2+j3-1)/j3-(y1-y0+1)/32;
if (vvv&1)
fprintf(stderr, "\n# lower serif x0,y0,j1-x0+x,y1-y %4d %4d %2d+%2d %2d",
x0,y0,j1-x0,x,y1-y);
/* check if really melted serifs */
if( loop(pp,j1,y,x1-x0,cs,0,RI)<1 ) { j1+=x; continue; }
if(num_cross(j1 ,j1+x,y,y,pp,cs) < 2 ){ j1+=x;continue; }
if (vvv&1) fprintf(stderr, " ok1");
j2 = j1 + loop(pp,j1,y,x1-x0,cs,0,RI);
j2 = j2 + loop(pp,j2,y,x1-x0,cs,1,RI);
i3 = loop(pp,j2,y,x1-x0,cs,0,RI); if(i3<2){j1+=x;continue;}
j2 += i3/2;
j3 = j2 + loop(pp,j2,y ,x1-j2,cs,0,RI);
i3 = j2 + loop(pp,j2,y-1,x1-j2,cs,0,RI); if(i3>j3)j3=i3;
j3 = j3 + loop(pp,j3,y ,x1-j3,cs,1,RI);
i3 = loop(pp,j3,y,x1-j3,cs,0,RI);
if(i3<2 || j3>=j1+x){j1+=x;continue;}
j3 += i3/2;
/* y =y1-(y1-y0+1+4)/8; */
if(x>5)
{
i++; /* snip! */
for(i3=0;i3<(y1-y0+1+4)/8;i3++)
put(pp,j2,y1-i3,255,128+64); /* clear highest bit */
if(vvv&4){
fprintf(stderr,"\n");
out_x(box2);
fprintf(stderr,"# melted serifs corrected on %d %d j1=%d j3=%d",j2-x0,y-y0,j1-x0,j3-x0);
}
for(xb=0,xa=0;xa<(x1-x0+4)/8;xa++){ /* detect vertical gap */
if( loop(pp,j2-xa,y0,y1-y0,cs,0,DO) > (y1-y0+1)/2
&& loop(pp,j2,(y0+y1)/2,xa+1,cs,0,LE) >=xa ){ xb=-xa; break; }
if( loop(pp,j2+xa,y0,y1-y0,cs,0,DO) > (y1-y0+1)/2
&& loop(pp,j2,(y0+y1)/2,xa+1,cs,0,RI) >=xa ){ xb= xa; break; }
}
if( get_bw(j2 ,j2 ,(y0+y1)/2,y1,pp,cs,1) == 0
&& get_bw(j2+xb,j2+xb,y0,(y0+y1)/2,pp,cs,1) == 0 )
{ /* divide */
box3=malloc_box(box2);
box3->x1=j2-1;
box2->x0=j2; x1=box2->x1;
cut_box(box2); /* cut vectors outside the box */
cut_box(box3);
box3->num=JOB->res.numC;
list_ins(&(JOB->res.boxlist),box2,box3); JOB->res.numC++; ii++;
/* box3,box2 in correct order??? */
if(vvv&4) fprintf(stderr," => splitted");
j1=x0=box2->x0; x=0; /* hopefully ok, NMK */
}
}
j1+=x;
}
progress(ni++,pc);
} end_for_each(&(JOB->res.boxlist));
close_progress(pc);
if(vvv)fprintf(stderr," %3d cluster corrected, %d new boxes\n",i,ii);
return 0;
}
/* remove black borders often seen on bad scanned copies of books
- dust around the border
*/
int remove_rest_of_dust() {
int i1, i2, vvv = JOB->cfg.verbose, x0, x1, y0, y1, cnt=0;
struct box *box2, *box4;
progress_counter_t *pc = NULL;
i1 = i2 = 0; /* counter for removed boxes */
if (vvv)
fprintf(stderr, "# detect dust (avX,nC), ... ");
/* remove fragments from border */
for_each_data(&(JOB->res.boxlist)) {
box2 = (struct box *)list_get_current(&(JOB->res.boxlist));
if (box2->c == UNKNOWN) {
x0 = box2->x0; x1 = box2->x1;
y0 = box2->y0; y1 = box2->y1; /* box */
/* box in char ??? */
if ( 2 * JOB->res.numC * (y1 - y0 + 1) < 3 * JOB->res.sumY
&& ( y1 < box2->p->y/4 || y0 > 3*box2->p->y/4 ) /* not single line */
&& JOB->res.numC > 1 /* do not remove everything */
&& ( box2->m4 == 0 ) ) /* remove this */
{
JOB->res.numC--; /* ToDo: dont count tiny pixels */
/* ToDo: res.sumX,Y must also be corrected */
i1++;
list_del(&(JOB->res.boxlist), box2);
free_box(box2);
}
}
} end_for_each(&(JOB->res.boxlist));
pc = open_progress(JOB->res.boxlist.n,"remove_dust2");
for_each_data(&(JOB->res.boxlist)) {
box2 = (struct box *)list_get_current(&(JOB->res.boxlist));
progress(cnt++,pc);
if (box2->c == PICTURE) continue;
x0 = box2->x0; x1 = box2->x1;
y0 = box2->y0; y1 = box2->y1; /* box */
/* remove tiny box2 if to far away from bigger boxes */
/* ToDo: remove clouds of tiny pixels (count near small, compare with num bigger) */
/* 0.42: remove far away pixel? ToDo: do it at earlier? */
if (x1-x0+1<3 && y1-y0+1<3){
int xn, yn, xs, ys;
int found=0; /* nearest bigger box */
/* search near bigger box */
for_each_data(&(JOB->res.boxlist)) {
box4 = (struct box *)list_get_current(&(JOB->res.boxlist));
if (found || box4 == box2) continue;
if (box4->x1-box4->x0+1<3 && box4->y1-box4->y0+1<3) continue;
xs = box4->x1-box4->x0+1;
ys = box4->y1-box4->y0+1;
xn = abs((box4->x0+box4->x1)/2 - box2->x0);
yn = abs((box4->y0+box4->y1)/2 - box2->y0);
if (2*xn < 3*xs && 2*yn < 3*ys) { found=1; }
} end_for_each(&(JOB->res.boxlist));
if (!found) { /* found nothing, box2 to far from big boxes */
i2++;
list_del(&(JOB->res.boxlist), box2);
free_box(box2);
}
}
} end_for_each(&(JOB->res.boxlist));
close_progress(pc);
if (vvv)
fprintf(stderr, " %3d + %3d boxes deleted, nC= %d ?\n",
i1, i2, JOB->res.numC);
return 0;
}