/* This is a Optical-Character-Recognition program Copyright (C) GPLv2 2000-2013 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 #include #include "gocr.h" #include "pnm.h" #include "pgm2asc.h" #include "unicode_defs.h" /* macro UNKNOWN */ #include #include #define Blen 256 // load boxes from database into boxlist (for faster access) // used as alternate engine, comparing chars with database // uses readpnm() and would conflict with multi images int load_db(job_t *job) { // called by gocr.c main() FILE *f1; char s1[Blen+1], s2[Blen+1] = "./db/", /* ToDo: replace by constant! by configure */ *s3; int i, j, ii, i2, line; struct box *box1; pix *pp; if( job->cfg.db_path ) strncpy(s2, job->cfg.db_path, Blen-1); i2=strlen(s2); if (job->cfg.verbose) fprintf(stderr, "# load database %s %s ... ", s2, job->cfg.db_path); strncpy(s2+i2,"db.lst",Blen-i2);s2[Blen]=0; f1 = fopen(s2, "r"); if (!f1) { fprintf(stderr, " DB %s not found\n",s2); return 1; } line = 0; /* line counter for better error report */ for (ii = 0; !feof(f1); ii++) { /* bbg: should write a better input routine */ if (!fgets(s1, Blen, f1)) break; line++; j = strlen(s1); /* remove carriage return sequences from line */ while (j > 0 && (s1[j - 1] == '\r' || s1[j - 1] == '\n')) s1[--j] = 0; if (!j) continue; /* skip empty line */ if (s1[0]=='#') continue; /* skip comments (v0.44) */ /* copy file name */ for (i = 0; i < j && i+i2 < Blen && strchr(" \t,;",s1[i]) == 0; i++) s2[i2 + i] = s1[i]; s2[i2+i]=0; /* skip spaces */ for (; i < j && strchr(" \t",s1[i]) != 0; i++); /* by now: read pix, fill box, goto next ??? */ pp = (pix *)malloc(sizeof(pix)); if( !pp ) fprintf(stderr,"malloc error in load_db pix\n"); // if (job->cfg.verbose) fprintf(stderr,"\n# readpgm %s ",s2); if (readpgm(s2, pp, 0 * job->cfg.verbose)!=0) { fprintf(stderr,"\ndatabase error: readpgm %s\n", s2); exit(-1); } box1 = (struct box *)malloc_box(NULL); if(!box1) fprintf(stderr,"malloc error in load_db box1\n"); box1->x0 = 0; box1->x1 = pp->x-1; // white border 1 pixel width box1->y0 = 0; box1->y1 = pp->y-1; box1->x = 1; box1->y = 1; box1->dots = 0; box1->c = 0; box1->modifier = 0; /* ToDo: obsolete */ box1->tas[0]=NULL; box1->tac[0]=0; box1->wac[0]=100; /* really 100% sure? */ box1->num_ac=1; if (s1[i]=='"'){ /* parse a string */ j=strrchr(s1+i+1,'"')-(s1+i+1); /* we only look for first and last "" */ if (j>=1) { s3=(char *)malloc(j+1); if (!s3) fprintf (stderr, "malloc error in load_db s3\n"); if (s3) { memcpy(s3,s1+i+1,j); s3[j]=0; box1->tas[0]=s3; // fprintf(stderr,"\nstring=%s",s3); } } else { fprintf(stderr,"load_db: string parse error L%d\n",line); } } else { box1->tac[0] = box1->c = s1[i]; /* try to interpret as ASCII */ /* we can live without hexcode in future if we use UTF8-strings */ s3=s1+i; j=strtol( s1+i, &s3, 16); /* try to read 4 to 8 digit hex unicode */ /* if its an hexcode, ASCII interpretation is overwritten */ if( j && i+3<=Blen && s3-s1-i>3 ) box1->tac[0] = box1->c = j; // fprintf(stderr,"\nhexcode=%04x=%04x %d",(int)j,(int)box1->c,s3-s1-i); } box1->num = 0; box1->line = -1; box1->m1 = 0; /* ToDo: should be given too in the database! */ box1->m2 = 0; box1->m3 = 0; box1->m4 = 0; box1->p = pp; list_app(&job->tmp.dblist, box1); // append to list #if 0 out_x(box1); #endif } fclose(f1); if (job->cfg.verbose) fprintf(stderr, " %d chars loaded\n", ii); return 0; } // expand database from box/boxlist name=db_$utime.pbm // this is added in version v0.3.3 int store_db(struct box *box1, job_t *job) { FILE *f1; char s2[Blen+1] = "./db/", s3[Blen+1]; int i2, dx, dy; unsigned c_out; pix b; /* temporary mini page */ if( job->cfg.db_path ) strncpy(s2,job->cfg.db_path,Blen-1); i2=strlen(s2); /* add (first) char and time to the file name for better debugging */ /* decide between 7bit ASCII and UTF8-char or string */ c_out = ((box1->num_ac && box1->tas[0]) ? (unsigned char )box1->tas[0][0] /* char */ : box1->c /* wchar */); /* (unsigned int)(( char)0x80) = 0xffffff80 */ /* (unsigned int)((unsigned char)0x80) = 0x00000080 */ /* name generation can cause problems, if called twice within a second */ sprintf(s3,"db_%04x_%08lx.pbm", c_out, (unsigned long)time(NULL)); /* ToDo: the file name may be not unique */ strncpy(s2+i2,"db.lst",Blen-i2);s2[Blen]=0; f1 = fopen(s2, "a"); if (!f1) { fprintf(stderr, " could not access %s\n",s2); return 1; } strncpy(s2+i2,s3,strlen(s3)); s2[i2+strlen(s3)]=0; /* store image and infos about the char */ /* ToDo: store the vector list instead of the pixelarray */ if (job->cfg.verbose) fprintf(stderr, "store_db: add file %s to database (nac=%d c=%04x)" "\n#",s3, box1->num_ac, c_out); dx=box1->x1-box1->x0+1; dy=box1->y1-box1->y0+1; b.p = (unsigned char *) malloc( dx * dy ); if( !b.p ){ fprintf( stderr, "\nFATAL: malloc failed, skip store_db" ); fclose(f1); return 2; } if (copybox(box1->p, box1->x0, box1->y0, dx, dy, &b, dx * dy)) { fclose(f1); return -1; } writepbm(s2,&b); /* What is to do on error? */ free(b.p); /* store the database line */ /* some infos about box1->m1,..,m4 should added (base line, high etc.) */ if (box1->num_ac && box1->tas[0]) { fprintf(f1, "%s \"%s\"\n",s3,box1->tas[0]); /* ToDo: what if tas contains '"'? */ } else { if( (box1->c >= '0' && box1->c <= '9') || (box1->c >= 'A' && box1->c <= 'Z') || (box1->c >= 'a' && box1->c <= 'z') ) fprintf(f1, "%s %c\n",s3,(char)box1->c); else { if (((box1->c)>>16)>>16) fprintf(f1, "%s %08x\n",s3,(unsigned int)box1->c); else fprintf(f1, "%s %04x\n",s3,(unsigned int)box1->c); } } fclose(f1); return 0; } /* function is only for user prompt on console to identify chars it prints out a part of pixmap b at point x0,y0 to stderr using dots .,; if no pixel, and @xoO for pixels */ void out_env(struct box *px, job_t *job){ int x0,y0,x1,y1,dx,dy,x,y,x2,y2,yy0,tx,ty,i,cs; char c1, c2; pix *b; cs=job->cfg.cs; yy0=px->y0; { /* overwrite rest of arguments */ b=px->p; x0=px->x0; x1=px->x1; dx=x1-x0+1; y0=px->y0; y1=px->y1; dy=y1-y0+1; y0-=2; y1+=2; if (px->m4 && y0>px->m1) y0=px->m1; if (px->m4 && y1m4) y1=px->m4; if (x1-x0+1<52) { x0-=10; x1+=10; } /* fragment? expand frame */ if (x1-x0+1<52) { x0-=10; x1+=10; } /* fragment? expand frame */ if (x1-x0+1<62) { x0-=5; x1+=5; } if (y1-y0+1<10) { y0-= 4; y1+= 4; } /* fragment? */ if (x0<0) x0=0; if (x1>=b->x) x1=b->x-1; if (y0<0) y0=0; if (y1>=b->y) y1=b->y-1; dx=x1-x0+1; dy=y1-y0+1; yy0=y0; fprintf(stderr,"\n# show box + environment"); fprintf(stderr,"\n# show box x= %4d %4d d= %3d %3d r= %d %d", px->x0, px->y0, px->x1 - px->x0 + 1, px->y1 - px->y0 + 1, px->x - px->x0, px->y - px->y0); if (px->num_ac){ /* output table of chars and its probabilities */ fprintf(stderr,"\n# list box char: "); for(i=0;inum_ac && itas[i]) fprintf(stderr," %s(%d)", px->tas[i] ,px->wac[i]); else fprintf(stderr," %s(%d)",decode(px->tac[i],ASCII),px->wac[i]); } fprintf(stderr,"\n"); if (px->dots && px->m2 && px->m1m1; dy=px->y1-yy0+1; } } tx=dx/80+1; ty=dy/40+1; // step, usually 1, but greater on large maps fprintf(stderr,"# show pattern x= %4d %4d d= %3d %3d t= %d %d\n", x0,y0,dx,dy,tx,ty); if (dx>0) for(y=yy0;yx0 || x > px->x1 || y+ty-1 < px->y0 || y > px->y1) c1=((c1=='#')?'O':','); fprintf(stderr,"%c", c1 ); } c1=c2=' '; /* mark lines with < */ if (px) if (y==px->m1 || y==px->m2 || y==px->m3 || y==px->m4) c1='<'; if (y==px->y0 || y==px->y1) c2='-'; /* boxmarks */ fprintf(stderr,"%c%c\n",c1,c2); } } /* // second variant, for database (with slightly other behaviour) // new variant // look at the environment of the pixel too (contrast etc.) // detailed analysis only of diff pixels! // // 100% * distance, 0 is best fit // = similarity of 2 chars for recognition of noisy chars // weigth of pixels with only one same neighbour set to 0 // look at contours too! ToDo: especially on small boxes distance should only be 0 if characters are 100% identical! */ // #define DEBUG 2 int distance2( pix *p1, struct box *box1, pix *p2, struct box *box2, int cs, int vvv){ int rc=0,x,y,v1,v2,i1,i2,rgood=0,rbad=0, x1,y1,x2,y2,dx,dy,dx1,dy1,dx2,dy2,tx,ty; #if DEBUG == 2 if (vvv) fprintf(stderr," DEBUG: distance2\n"); #endif x1=box1->x0;y1=box1->y0;x2=box2->x0;y2=box2->y0; dx1=box1->x1-box1->x0+1; dx2=box2->x1-box2->x0+1; dx=((dx1>dx2)?dx1:dx2);dx=dx1; dy1=box1->y1-box1->y0+1; dy2=box2->y1-box2->y0+1; dy=((dy1>dy2)?dy1:dy2);dy=dy1; if(abs(dx1-dx2)>1+dx/16 || abs(dy1-dy2)>1+dy/16) rbad++; // how to weight? // compare relations to baseline and upper line if(box1->m4>0 && box2->m4>0){ // used ??? if(2*box1->y1>box1->m3+box1->m4 && 2*box2->y1m3+box2->m4) rbad+=128; if(2*box1->y0>box1->m1+box1->m2 && 2*box2->y0m1+box2->m2) rbad+=128; } tx=dx/16; if(dx<17)tx=1; // raster ty=dy/32; if(dy<33)ty=1; // compare pixels for( y=0;y0) rbad+=16*v1; } if(rgood+rbad) rc= 100*rbad/(rgood+rbad); else rc=99; /* if width/high is not correct add badness */ rc += ( abs(dx1*dy2-dx2*dy1) * 10 ) / (dy1*dy2); if (rc>100) rc=100; if(/* rc<10 && */ vvv /* &1024 */){ #if DEBUG == 2 fprintf(stderr," distance2 rc=%d rgood=%d rbad=%d\n",rc,rgood,rbad); // out_b(NULL,p1,box1->x0,box1->y0,box1->x1-box1->x0+1, // box1->y1-box1->y0+1,cs); // out_b(NULL,p2,box2->x0,box2->y0,box2->x1-box2->x0+1, // box2->y1-box2->y0+1,cs); out_x(box1); out_x(box2); #endif } return rc; } wchar_t ocr_db(struct box *box1, job_t *job) { int dd = 1000, dist = 1000; wchar_t c = UNKNOWN; unsigned char buf[200]; /* Oct08 JS: add unsigned to avoid UTF problems */ Box *box2, *box3; if (!list_empty(&job->tmp.dblist)){ box3 = (Box *)list_get_header(&job->tmp.dblist); if(job->cfg.verbose) fprintf(stderr,"\n#DEBUG: ocr_db (%d,%d) ",box1->x0, box1->y0); for_each_data(&job->tmp.dblist) { box2 = (Box *)list_get_current(&job->tmp.dblist); /* do preselect!!! distance() slowly */ dd = distance2( box2->p, box2, box1->p, box1, job->cfg.cs, job->cfg.verbose); if (dd <= dist) { /* new best fit */ dist = dd; box3 = box2; /* box3 is a pointer and not copied box2 */ if (dist<100 && 100-dist >= job->cfg.certainty) { /* some deviation of the pattern is tolerated */ int i, wa; for (i=0;inum_ac;i++) { wa = (100-dist)*box3->wac[i]/100; /* weight *= (100-dist) */ if (box3->tas[i]) setas(box1,box3->tas[i],wa); else setac(box1,box3->tac[i],wa); } if (box3->num_ac) c=box3->tac[0]; /* 0 for strings (!UNKNOWN) */ if (job->cfg.verbose) fprintf(stderr, " dist=%4d c= %c 0x%02x %s wc= %3d", dist, ((box3->c>32 && box3->c<127) ? (char) box3->c : '.'), (int)box3->c, ((box3->tas[0])?box3->tas[0]:""), box3->wac[0]); } if (dd<=0 && ((box3->num_ac && box3->tas[0]) || box3->c >= 128 || !strchr ("l1|I0O", box3->c))) break; /* speedup if found */ } } end_for_each(&job->tmp.dblist); } if( (job->cfg.mode&128) != 0 && c == UNKNOWN ) { /* prompt the user */ /* should the output go to stderr or special pipe??? */ int utf8_ok=0; /* trigger this flag if input is ok */ int i, endchar; /* index */ out_env(box1, job); /* old: out_x(box1); */ fprintf(stderr,"The above pattern was not recognized.\n" "Enter UTF8 char or string for above pattern. Leave empty if unsure.\n" "Press RET at the end (ALT+RET to store into RAM only) : " ); /* ToDo: empty + alt-return (0x1b 0x0a) for help? ^a for skip all */ /* UTF-8 (man 7 utf-8): * 7bit = 0xxxxxxx (0000-007F) * 11bit = 110xxxxx 10xxxxxx (0080-07FF) * 16bit = 1110xxxx 10xxxxxx 10xxxxxx (0800-FFFF) * 21bit = 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx * 26bit = 111110xx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx * 31bit = 1111110x 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx */ buf[0]=0; /* shift/ctrl/altgr-enter acts like enter or ^j or ^m, * alt-enter returns 0x1b 0x0a and returns from fgets() * ^d (EOF) returns (nil) from fgets() * x+(2*)ctrl-d returns from fgets() without returning a 0x0a * if not UTF-input-mode, we are in trouble? * ^a=0x01, ^b=0x02, ^e=05, ..., ToDo: meaning of no-input or <=space */ fgets((char *)buf,200,stdin); /* including \n=0x0a */ dd=strlen((char *)buf); /* output hexcode if verbose set */ if (job->cfg.verbose) { fprintf(stderr, "\n# fgets [%d]:", dd); for(i=0; icfg.mode&=~128; } /* skip all */ buf[dd=i]=0; /* replace final 0x0a or other special codes */ if (dd==1 && !(buf[0]&128)) { c=buf[0]; utf8_ok=1; } /* single char */ if (dd>1 && dd<7) { /* try to decode single wide char (utf8) */ int u0, u1; /* define UTF8-start sequences, u0=0bits u1=1bits */ u0= 1<<(7-dd); /* compute start byte from UTF8-length */ u1=255&~((1<<(8-dd))-1); /* count number of following 10xxxxxx bytes to i */ for (i=1;i0){ /* ToDo: skip space and tab too? */ if (utf8_ok==1) { setac(box1, c, 100); } /* store single wchar */ if (utf8_ok==0) { /* store a string of chars (UTF8-string) */ c='_'; /* what should we do with c? probably a bad idea? */ setas(box1, (char *)buf, 100); } /* decide between * 0) just help gocr to find the results and (dont remember, 0x01) * 1) help and remember in the same run (store to memory, 0x1b) * 2) expand the database (dont store ugly chars to the database!) */ if (endchar!=0x01){ /* ^a before hit return */ /* is there a reason to dont store to memory? */ list_app(&job->tmp.dblist, box1); /* append to list for 1+2 */ } if (endchar!=0x01 && endchar!=0x1b){ store_db(box1, job); /* store to disk for 2 */ } if (job->cfg.verbose) fprintf(stderr, " got char= %c 16bit= 0x%04x string= \"%s\"\n", ((c>32 && c<127)?(char)c:'.'), (int)c, buf); } } return c; }