+++ /dev/null
-/*
- * Core OCR algorithm (first exact bitmap match)
- */
-/*
- * This is part of ypp-sc-tools, a set of third-party tools for assisting
- * players of Yohoho Puzzle Pirates.
- *
- * Copyright (C) 2009 Ian Jackson <ijackson@chiark.greenend.org.uk>
- *
- * 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 3 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, see <http://www.gnu.org/licenses/>.
- *
- * Yohoho and Puzzle Pirates are probably trademarks of Three Rings and
- * are used without permission. This program is not endorsed or
- * sponsored by Three Rings.
- */
-
-#include "ocr.h"
-#include "convert.h"
-
-typedef struct {
- Pixcol col;
- struct DatabaseNode *then;
-} DatabaseLink;
-
-typedef struct DatabaseNode {
- char *str;
- int nlinks, alinks;
- unsigned match:1, defined:1, endsword:1, local:1;
- DatabaseLink *links;
-} DatabaseNode;
-
-typedef struct {
- OcrReader *rd;
- OcrCellType ct;
- int w;
- const Pixcol *cols;
- int x;
- unsigned ctxmap;
-} FindCharArgs;
-typedef struct {
- DatabaseNode *match;
- int rx;
-} FindCharResults;
-
-#define FOR_EACH_CONTEXT(EACH) \
- EACH(Word) \
- EACH(Lower) \
- EACH(Upper) \
- EACH(Digit)
-
-#define FEC_ENUM(Context) ct_##Context,
-#define FEC_BIT(Context) ctf_##Context = 1 << ct_##Context,
-enum {
- FOR_EACH_CONTEXT(FEC_ENUM)
- FOR_EACH_CONTEXT(FEC_BIT)
-};
-
-#define FEC_STRINGS(Context) #Context,
-static const char *context_names[]= { FOR_EACH_CONTEXT(FEC_STRINGS) };
-
-struct OcrCellTypeInfo {
- /* bitmaps of indices into context_names: */
- unsigned initial, nextword, midword;
- int space_spaces;
- const char *name;
- int (*findchar_select)(const FindCharArgs *fca,
- const FindCharResults results[]);
-};
-
-#define NCONTEXTS (sizeof(context_names)/sizeof(context_names[0]))
-
-struct OcrReader {
- int h;
- DatabaseNode contexts[NCONTEXTS];
- char *result;
- int lresult, aresult;
- OcrResultGlyph *results;
- int aresults, nresults;
-};
-
-DEBUG_DEFINE_DEBUGF(ocr)
-
-#define FGETSLINE (dbfile_getsline(lbuf,sizeof(lbuf),__FILE__,__LINE__))
-
-static void cleardb_node(DatabaseNode *n) {
- int i;
- free(n->str); n->str=0;
- n->defined=n->match=n->endsword= 0;
- for (i=0; i<n->nlinks; i++)
- cleardb_node(n->links[i].then);
-}
-
-static void readdb1(OcrReader *rd, const char *which, int local);
-
-static void readdb(OcrReader *rd) {
- int ctxi;
-
- for (ctxi=0; ctxi<NCONTEXTS; ctxi++)
- cleardb_node(&rd->contexts[ctxi]);
-
- readdb1(rd, "master", 0);
- readdb1(rd, "local", 1);
-}
-
-static void readdb1(OcrReader *rd, const char *which, int local) {
- int nchrs;
- DatabaseNode *current, *additional;
- char chrs[100];
- Pixcol cv;
- int j,ctxi;
- int h, endsword;
- char lbuf[100];
-
- char *dbfname= masprintf("%s/_%s-char%d.txt",
- get_vardir(), which, rd->h);
-
- if (!dbfile_open(dbfname))
- goto x;
-
- if (local)
- o_flags |= ff_charset_havelocal;
-
- FGETSLINE;
- dbassert(!strcmp(lbuf,"# ypp-sc-tools pctb font v3 depth=" STRING(AADEPTH)));
-
- dbassert( dbfile_scanf("%d", &h) == 1);
- dbassert(h==rd->h);
-
- for (;;) {
- FGETSLINE;
- if (!lbuf[0] || lbuf[0]=='#') continue;
- if (!strcmp(lbuf,".")) break;
-
- for (ctxi=0; ctxi<NCONTEXTS; ctxi++)
- if (!strcmp(lbuf,context_names[ctxi]))
- goto found_ctx;
- /* not found, just skip */
- for (;;) { FGETSLINE; if (!lbuf[0]) break; }
- continue;
-
- found_ctx:
- for (nchrs=0;;) {
- int c= fgetc(dbfile); sysassert(!ferror(dbfile)); dbassert(c!=EOF);
- if (c=='\n') break; /* forces no match */
- dbassert(nchrs<sizeof(chrs));
- chrs[nchrs++]= c;
- }
- endsword= 0;
- if (nchrs>0 && chrs[nchrs-1]==' ') {
- endsword= 1;
- nchrs--;
- }
-
- current= &rd->contexts[ctxi];
- for (;;) {
- FGETSLINE;
- if (!lbuf[0]) { dbassert(current != &rd->contexts[ctxi]); break; }
- dbassert( strlen(lbuf) == rd->h );
- FILLZERO(cv);
- int y;
- for (y=0; y<h; y++) {
- char pb[2];
- pb[0]= lbuf[y];
- pb[1]= 0;
- char *ep;
- unsigned pv= strtoul(pb,&ep,16); dbassert(!*ep);
- pixcol_p_add(&cv, y, pv);
- }
-
- for (j=0; j<current->nlinks; j++)
- if (!pixcol_cmp(&cv, ¤t->links[j].col)) {
- current= current->links[j].then;
- goto found_link;
- }
-
- additional= mmalloc(sizeof(*additional));
- additional->str= 0;
- additional->defined= 0;
- additional->match= 0;
- additional->endsword= 0;
- additional->nlinks= additional->alinks= 0;
- additional->links= 0;
- if (current->nlinks==current->alinks) {
- current->alinks++;
- current->alinks<<=1;
- current->links= mrealloc(current->links,
- sizeof(*current->links) * current->alinks);
- }
- current->links[current->nlinks].col= cv;
- current->links[current->nlinks].then= additional;
- current->nlinks++;
- current= additional;
-
- found_link:;
- }
-
- if (!current->defined) {
- free(current->str);
- current->str= 0;
- current->defined= 1;
- current->match= 0;
- current->local= local;
-
- if (nchrs) {
- current->str= mmalloc(nchrs+1);
- memcpy(current->str, chrs, nchrs);
- current->str[nchrs]= 0;
- current->match= 1;
- current->endsword= endsword;
- }
- }
- }
- x:
- dbfile_close();
- free(dbfname);
-}
-
-typedef struct Rejection Rejection;
-struct Rejection {
- struct Rejection *next;
- const char *fname;
- int lno;
- pcre *re;
-};
-
-Rejection *rejections;
-
-static void load_rejections(const char *which) {
- char lbuf[1000];
- char *fname= masprintf("%s/_%s-reject.txt", get_vardir(), which);
- int c, lno=0;
- Rejection *rej;
-
- if (!dbfile_open(fname)) { free(fname); return; }
-
- while ((c= fgetc(dbfile))!=EOF) {
- ungetc(c,dbfile);
- lno++;
- dbfile_getsline(lbuf,sizeof(lbuf),fname,lno);
-
- if (!lbuf[0] || isspace(lbuf[0] || lbuf[0]=='#'))
- continue;
-
- rej= mmalloc(sizeof(*rej));
- rej->next= rejections;
- rej->fname= fname;
- rej->lno= lno;
-
- const char *err;
- int erroffset;
- rej->re= pcre_compile(lbuf, PCRE_NO_AUTO_CAPTURE|PCRE_UTF8,
- &err, &erroffset, 0);
- if (!rej->re) {
- char *what= masprintf("invalid regexp at offset %d: %s\n",
- erroffset, err);
- dbfile_assertfail(fname, lno, what);
- }
- debugf("OCR LOADED REJECTION %s:%d `%s' %p\n", fname,lno, lbuf, rej->re);
-
- rejections= rej;
- }
- sysassert(feof(dbfile));
- dbfile_close();
-}
-
-static int should_reject(OcrReader *rd) {
- static int rejections_loaded;
- int ovector[30];
-
- if (!rejections_loaded) {
- fetch_with_rsync("reject");
- load_rejections("master");
- load_rejections("local");
- rejections_loaded=1;
- }
-
- debugf("[OCR REJECTION `%s'%d", rd->result, rd->lresult);
- Rejection *rej;
- for (rej=rejections; rej; rej=rej->next) {
- debugf(" (%p)", rej);
-
- int res= pcre_exec(rej->re, 0, rd->result,rd->lresult, 0,
- 0, ovector, ARRAYSIZE(ovector));
- if (res==PCRE_ERROR_NOMATCH) continue;
- sysassert(res>0);
-
- debugf(" MATCH]\n");
- fprintf(stderr,"Rejecting OCR result `%s' (due to %s:%d)\n",
- rd->result, rej->fname,rej->lno);
- return 1;
- }
- debugf(" OK]");
- return 0;
-}
-
-static void cu_pr_ctxmap(FILE *resolver, unsigned ctxmap) {
- fprintf(resolver,"{");
- const char *spc="";
- int ctxi;
- for (ctxi=0; ctxi<NCONTEXTS; ctxi++) {
- if (!(ctxmap & (1u << ctxi))) continue;
- fprintf(resolver,"%s%s",spc,context_names[ctxi]);
- spc=" ";
- }
- fprintf(resolver,"}");
-}
-
-static void callout_unknown(OcrReader *rd, int w, const Pixcol cols[],
- int unk_l, int unk_r, unsigned unk_ctxmap) {
- int c,i, x,y;
- const OcrResultGlyph *s;
- const char *p;
-
- FILE *resolver= resolve_start();
- if (!resolver ||
- !((o_flags & ff_charset_edit) ||
- ((o_flags & ff_charset_allowedit) &&
- (o_flags & ff_charset_havelocal))))
- fatal("OCR failed - unrecognised characters or ligatures.\n"
- "Character set database needs to be updated or augmented.\n"
- "See README.charset.\n");
-
- fprintf(resolver,
- "char\n"
- "%d %d ",unk_l,unk_r);
- cu_pr_ctxmap(resolver,unk_ctxmap);
- for (i=0, s=rd->results; i<rd->nresults; i++, s++) {
- if (!strcmp(s->s," ")) continue;
- fprintf(resolver," %d %d ",s->l,s->r);
- cu_pr_ctxmap(resolver, 1u << s->match);
- fprintf(resolver," ");
- cu_pr_ctxmap(resolver, s->ctxmap);
- fprintf(resolver," ");
- for (p=s->s; (c= *p); p++) {
- if (c=='\\') fprintf(resolver,"\\%c",c);
- else if (c>=33 && c<=126) fputc(c,resolver);
- else fprintf(resolver,"\\x%02x",(unsigned char)c);
- }
- }
- fputc('\n',resolver);
-
- fprintf(resolver,
- "P2\n%d %d %d\n", w, rd->h, AAMAXVAL);
- for (y=0; y<rd->h; y++) {
- for (x=0; x<w; x++)
- fprintf(resolver, " %d", pixcol_p_get(&cols[x], y));
- fputc('\n',resolver);
- }
-
- resolve_finish();
- readdb(rd);
-}
-
-static void add_result(OcrReader *rd, const char *s, int l, int r,
- int match, unsigned ctxmap) {
- if (rd->nresults >= rd->aresults) {
- rd->aresults++; rd->aresults<<=1;
- rd->results= mrealloc(rd->results, sizeof(*rd->results)*rd->aresults);
- }
- rd->results[rd->nresults].s= s;
- rd->results[rd->nresults].l= l;
- rd->results[rd->nresults].r= r;
- rd->results[rd->nresults].match= match;
- rd->results[rd->nresults].ctxmap= ctxmap;
- rd->nresults++;
-
- if (!s) return; /* just the sentinel for the caller */
-
- int sl= strlen(s);
- int newlresult= rd->lresult + sl;
- if (newlresult >= rd->aresult) {
- rd->aresult= (newlresult << 1) + 1;
- rd->result= mrealloc(rd->result, rd->aresult);
- }
- memcpy(rd->result + rd->lresult, s, sl);
- rd->lresult= newlresult;
- rd->result[rd->lresult]= 0;
-}
-
-
-static DatabaseNode *findchar_1ctx(const FindCharArgs *fca,
- DatabaseNode *start, int *matchx_r) {
- DatabaseNode *current= start;
- DatabaseNode *bestmatch= 0;
- int i;
- int x= fca->x;
-
- for (;;) {
- if (DEBUGP(ocr)) debug_flush();
- debugf(" | x=%d",x);
- if (x > fca->w) break;
- Pixcol cv= fca->cols[x];
- debugf(" cv="PIXCOL_PRFMT, PIXCOL_PRVAL(cv));
- for (i=0; i<current->nlinks; i++)
- if (!pixcol_cmp(&cv, ¤t->links[i].col))
- goto found;
- /* not found */
- debugf(" ?");
- break;
-
- found:
- current= current->links[i].then;
- if (current->match) {
- debugf(" \"%s\"%s",current->str,current->endsword?"_":"");
- bestmatch= current;
- *matchx_r= x;
- } else {
- debugf(" ...");
- }
-
- x++;
- }
- return bestmatch;
-}
-
-static DatabaseNode *findchar(const FindCharArgs *fca,
- int *match_rx, int *match_rctxi) {
- FindCharResults results[NCONTEXTS];
- int ctxi, match=-1, nmatches=0;
-
- debugf("OCR lx=%d ct_state=%x ", fca->x, fca->ctxmap);
- for (ctxi=0; ctxi<NCONTEXTS; ctxi++) {
- results[ctxi].match= 0;
- if (!(fca->ctxmap & (1u << ctxi))) continue;
- debugf(" || %s",context_names[ctxi]);
-
- results[ctxi].match= findchar_1ctx(fca, &fca->rd->contexts[ctxi],
- &results[ctxi].rx);
- if (!results[ctxi].match) continue;
-
- match= ctxi;
- nmatches++;
- }
- if (nmatches==1) {
- debugf(" unique");
- } else {
- debugf(" ambiguous");
- match= !fca->ct->findchar_select ? -1 :
- fca->ct->findchar_select(fca,results);
- debugf(" resolved %s", match<0 ? "<none>" : context_names[match]);
- }
- if (match<0)
- return 0;
-
- *match_rx= results[match].rx;
- if (match_rctxi) *match_rctxi= match;
- return results[match].match;
-}
-
-static int findchar_select_text(const FindCharArgs *fca,
- const FindCharResults results[]) {
-
- dbassert(! results[ct_Digit].match ); /* digits are supposedly unambiguous */
-
- switch (fca->ctxmap) {
-
-#define RETURN_IF_LONGER(this,that) do{ \
- if (results[ct_##this].rx > results[ct_##that].rx) \
- return ct_##this; \
- }while(0)
-
- case ctf_Digit | ctf_Upper | ctf_Lower | ctf_Word:
- /* Start of word. Prefer Word match; failing that, take the longest */
- if (results[ct_Word].match) return ct_Word;
- RETURN_IF_LONGER(Lower,Upper);
- RETURN_IF_LONGER(Upper,Lower);
- break;
-
- case ctf_Digit | ctf_Upper | ctf_Lower:
- /* Mid-word. Prefer longer match; failing that, match lower. */
- RETURN_IF_LONGER(Upper,Lower);
- return ct_Lower;
- }
-
- /* oh well */
- return -1;
-}
-
-const struct OcrCellTypeInfo ocr_celltype_number= {
- ctf_Digit, ctf_Digit, ctf_Digit,
- .space_spaces= 5,
- .name= "number",
- .findchar_select= 0
-};
-const struct OcrCellTypeInfo ocr_celltype_text= {
- .initial= ctf_Digit | ctf_Upper,
- .nextword= ctf_Digit | ctf_Upper | ctf_Lower | ctf_Word,
- .midword= ctf_Digit | ctf_Upper | ctf_Lower,
- .space_spaces= 4,
- .name= "text",
- .findchar_select= findchar_select_text
-};
-
-
-const char *ocr_celltype_name(OcrCellType ct) { return ct->name; }
-
-OcrResultGlyph *ocr(OcrReader *rd, OcrCellType ct, int w,
- const Pixcol cols[]) {
- int nspaces;
- int x;
-
- FindCharArgs fca;
- fca.rd= rd;
- fca.ct= ct;
- fca.w= w;
- fca.cols= cols;
- fca.x= -1;
-
- restart:
-
- nspaces=- w;
- fca.ctxmap= ct->initial;
- rd->nresults=0;
- rd->lresult=0;
- rd->result[0]=0;
- debugf("OCR h=%d w=%d",rd->h,w);
- for (x=0; x<w; x++) debugf(" "PIXCOL_PRFMT, PIXCOL_PRVAL(cols[x]));
- debugf("\n");
- debug_flush();
-
- x=0;
- for (;;) {
- if (DEBUGP(ocr)) debug_flush();
- /* skip spaces */
- if (x>=w)
- break;
-
- if (!pixcol_nonzero(&cols[x])) {
- nspaces++;
- x++;
- if (nspaces == ct->space_spaces) {
- debugf("OCR x=%x nspaces=%d space\n",x,nspaces);
- fca.ctxmap= ct->nextword;
- }
- continue;
- }
-
- /* something here, so we need to add the spaces */
- if (nspaces >= ct->space_spaces)
- add_result(rd," ",x-nspaces,x+1,-1,0);
- nspaces=0;
-
- fca.x= x;
-
- int match_rx=-1;
- int match_ctxi=-1;
- DatabaseNode *match= findchar(&fca, &match_rx, &match_ctxi);
-
- if (match) {
- debugf(" || YES");
- add_result(rd, match->str, x, match_rx, match_ctxi, fca.ctxmap);
- if (should_reject(rd)) {
- callout_unknown(rd, w,cols, match_rx+1,match_rx, 0);
- goto restart;
- }
- x= match_rx+1;
- if (match->match) fca.ctxmap= ct->midword;
- else debugf(" (empty)");
- if (match->endsword) {
- nspaces= ct->space_spaces;
- debugf("_");
- fca.ctxmap= ct->nextword;
- }
- debugf("\n");
- } else {
- int rx;
- debugf(" || UNKNOWN");
- for (rx=x; rx<w && pixcol_nonzero(&cols[rx]); rx++);
- debugf(" x=%d ctxmap=%x %d..%d\n",x, fca.ctxmap, x,rx);
- debug_flush();
- callout_unknown(rd, w,cols, x,rx-1, fca.ctxmap);
- goto restart;
- }
-
- }
- add_result(rd, 0,-1,-1,-1,0);
- debugf("OCR finished %d glyphs\n",rd->nresults);
- debug_flush();
- return rd->results;
-}
-
-OcrReader *ocr_init(int h) {
- OcrReader *rd;
-
- assert(h <= OCR_MAX_H);
-
- if (o_flags & ff_dict_fetch) {
- char *fetchfile= masprintf("char%d",h);
- fetch_with_rsync(fetchfile);
- free(fetchfile);
- }
-
- rd= mmalloc(sizeof(*rd));
- memset(rd,0,sizeof(*rd));
-
- rd->h= h;
-
- rd->aresult= 10;
- rd->result= mmalloc(rd->aresult);
-
- readdb(rd);
- return rd;
-}
-
-/*---------- character set dump ----------*/
-
-static void show_recurse(const DatabaseNode *t, int *count,
- const DatabaseNode **store_ary) {
- if (t->defined) {
- if (store_ary) store_ary[*count]= t;
- (*count)++;
- }
- int l;
- for (l=0; l<t->nlinks; l++)
- show_recurse(t->links[l].then, count,store_ary);
-}
-
-static int show_char_compar(const void *av, const void *bv) {
- const DatabaseNode *const *ap= av; const DatabaseNode *a= *ap;
- const DatabaseNode *const *bp= bv; const DatabaseNode *b= *bp;
- return strcmp(a->str, b->str) ?:
- ((int)a->match - (int)b->match) ?:
- ((int)a->endsword - (int)b->endsword) ?:
- ((int)a->local - (int)b->local) ?:
- 0;
-}
-
-void ocr_showcharsets(void) {
- DIR *d;
- struct dirent *de;
- char found[OCR_MAX_H];
- pcre *fnpat;
- int matchvec[10];
- char hbuf[10];
- const char *pcre_err;
- int pcre_erroffset;
-
- memset(found,0,sizeof(found));
-
- fnpat= pcre_compile("_(?:master|local)\\-char([1-9]\\d{0,2})\\.txt$",
- PCRE_ANCHORED|PCRE_DOLLAR_ENDONLY,
- &pcre_err,&pcre_erroffset, 0);
- debugf("pcre_compile %p %s\n",fnpat,pcre_err);
- assert(fnpat);
-
- sysassert( d= opendir(get_vardir()) );
- for (;;) {
- errno=0; de= readdir(d); if (!de) break;
-
- int rer= pcre_exec(fnpat,0, de->d_name,strlen(de->d_name), 0,0,
- matchvec,ARRAYSIZE(matchvec));
- debugf("pcre_exec `%s' => %d\n", de->d_name,rer);
-
- if (rer==PCRE_ERROR_NOMATCH || rer==PCRE_ERROR_BADUTF8) continue;
- assert(rer==2);
-
- rer= pcre_copy_substring(de->d_name,matchvec,rer, 1, hbuf,sizeof(hbuf));
- debugf("pcre_copy_substring => %d\n", rer);
- assert(rer>0);
-
- int h= atoi(hbuf);
- if (h >= ARRAYSIZE(found)) continue;
-
- found[h]= 1;
- }
-
- int h;
- for (h=0; h<ARRAYSIZE(found); h++) {
- if (!found[h]) continue;
-
- OcrReader *rd= ocr_init(h); /* we leak this but never mind */
- progress("");
-
- int ctxi;
- for (ctxi=0; ctxi<NCONTEXTS; ctxi++) {
- int nchars= 0;
- show_recurse(&rd->contexts[ctxi], &nchars, 0);
- const DatabaseNode **chars= mmalloc(sizeof(*chars) * nchars);
- int chari= 0;
- show_recurse(&rd->contexts[ctxi], &chari, chars);
- assert(chari==nchars);
- qsort(chars, nchars, sizeof(*chars), show_char_compar);
-
- int local;
- for (local=0; local<2; local++) {
- printf("%2d %-6s %-6s ", h, context_names[ctxi],
- local?"local":"master");
- for (chari=0; chari<nchars; chari++) {
- const DatabaseNode *t= chars[chari];
- static const char accept[]=
- "abcdefghijklmnopqrstuvwxyz"
- "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
- "0123456789" "'!&-+.,>";
-
- if (t->local != local) continue;
-
- if (!t->match)
- printf(" [nomatch]");
- else if (!t->endsword && strspn(t->str, accept) == strlen(t->str))
- printf(" %s",t->str);
- else {
- printf(" \"");
- char *p= t->str;
- int c;
- while ((c=*p++)) {
- if (c=='"' || c=='\\') printf("\\%c",c);
- else if (c>=' ' && c<=126) putchar(c);
- else printf("\\x%02x", (unsigned char)c);
- }
- if (t->endsword) putchar(' ');
- putchar('"');
- }
- }
- putchar('\n');
- }
- free(chars);
- }
- }
-}