X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~yarrgweb/git?a=blobdiff_plain;f=pctb%2Focr.c;h=6448ee44d5e72bb275e17459a2863e3be68d7d2c;hb=479a3cc7b4b79e8cb4df95664da5a9b0cb7918de;hp=ecfac976b2d025176bd4ee82675c31a108c3ba18;hpb=3a820e9871e6c4607b34939cc4d55c6925a7de29;p=ypp-sc-tools.db-test.git diff --git a/pctb/ocr.c b/pctb/ocr.c index ecfac97..6448ee4 100644 --- a/pctb/ocr.c +++ b/pctb/ocr.c @@ -40,31 +40,44 @@ typedef struct DatabaseNode { DatabaseLink *links; } DatabaseNode; -static const char *context_names[]= { - "Lower", - "Upper", - "Digit" +typedef struct { + OcrReader *rd; + OcrCellType ct; + int w; + Pixcol *cols; + int x; + unsigned ctxmap; +} FindCharArgs; +typedef struct { + DatabaseNode *match; + int rx; +} FindCharResults; + +#define FOR_EACH_CONTEXT(EACH) \ + EACH(Word) \ + EACH(Upper) \ + EACH(Lower) \ + 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; -}; -const struct OcrCellTypeInfo ocr_celltype_number= { - 4,4,4, - .space_spaces= 5, - .name= "number" -}; -const struct OcrCellTypeInfo ocr_celltype_text= { - .initial=2, /* Uppercase */ - .nextword=3, /* Either */ - .midword=1, /* Lower only */ - .space_spaces= 4, - .name= "text" + int (*findchar_select)(const FindCharArgs *fca, + const FindCharResults results[]); }; - #define NCONTEXTS (sizeof(context_names)/sizeof(context_names[0])) struct OcrReader { @@ -219,7 +232,7 @@ static void callout_unknown(OcrReader *rd, int w, Pixcol cols[], Pixcol pv; FILE *resolver= resolve_start(); - if (!resolver) + if (!resolver || !(o_flags & ff_editcharset)) fatal("OCR failed - unrecognised characters or ligatures.\n" "Character set database needs to be updated or augmented.\n" "See README.charset.\n"); @@ -231,7 +244,9 @@ static void callout_unknown(OcrReader *rd, int w, Pixcol cols[], for (i=0, s=rd->results; inresults; i++, s++) { if (!strcmp(s->s," ")) continue; fprintf(resolver," %d %d ",s->l,s->r); - cu_pr_ctxmap(resolver,s->ctxmap); + 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); @@ -262,7 +277,7 @@ static void callout_unknown(OcrReader *rd, int w, Pixcol cols[], } static void add_result(OcrReader *rd, const char *s, int l, int r, - unsigned ctxmap) { + int match, unsigned ctxmap) { if (rd->nresults >= rd->aresults) { rd->aresults++; rd->aresults<<=1; rd->results= mrealloc(rd->results, sizeof(*rd->results)*rd->aresults); @@ -270,24 +285,24 @@ static void add_result(OcrReader *rd, const char *s, int l, int r, 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++; } -const char *ocr_celltype_name(OcrCellType ct) { return ct->name; } - -static DatabaseNode *findchar(OcrReader *rd, int w, Pixcol cols[], - int x, int ctxi, int *matchx_r) { - DatabaseNode *current= &rd->contexts[ctxi]; +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 (;;) { debug_flush(); debugf(" | x=%d",x); - if (x>w) break; - Pixcol cv= cols[x]; + if (x > fca->w) break; + Pixcol cv= fca->cols[x]; debugf(" cv=%"PSPIXCOL(PRIx),cv); for (i=0; inlinks; i++) if (current->links[i].col == cv) @@ -311,15 +326,84 @@ static DatabaseNode *findchar(OcrReader *rd, int w, Pixcol cols[], 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; ctxictxmap & (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 ? "" : 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[]) { + if (fca->ctxmap != 017) return -1; + + dbassert(! results[ct_Digit].match ); + if (results[ct_Word].match) return ct_Word; + if (results[ct_Lower].rx > results[ct_Upper].rx) return ct_Lower; + if (results[ct_Upper].rx > results[ct_Lower].rx) return ct_Upper; + 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_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, Pixcol cols[]) { int nspaces; - unsigned ctxmap; - int ctxi, x; + int x; + + FindCharArgs fca; + fca.rd= rd; + fca.ct= ct; + fca.w= w; + fca.cols= cols; + fca.x= -1; restart: nspaces=- w; - ctxmap= ct->initial; + fca.ctxmap= ct->initial; rd->nresults=0; debugf("OCR h=%d w=%d",rd->h,w); for (x=0; xspace_spaces) { debugf("OCR x=%x nspaces=%d space\n",x,nspaces); - ctxmap= ct->nextword; + 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+1,0); + add_result(rd," ",x-nspaces,x+1,-1,0); nspaces=0; - DatabaseNode *uniquematch= 0; - int uniquematch_rx=-1; - - debugf("OCR lx=%d ctxmap=%x ",x,ctxmap); - - for (ctxi=0; ctxistr, uniquematch->str)) { - debugf( " ambiguous"); - uniquematch= 0; - } else { - uniquematch= match; - } - } - - if (uniquematch) { + int match_rx=-1; + int match_ctxi=-1; + DatabaseNode *match= findchar(&fca, &match_rx, &match_ctxi); + + if (match) { debugf(" || YES"); - add_result(rd, uniquematch->str, x, uniquematch_rx, ctxmap); - x= uniquematch_rx+1; - if (uniquematch->match) ctxmap= ct->midword; + add_result(rd, match->str, x, match_rx, match_ctxi, fca.ctxmap); + x= match_rx+1; + if (match->match) fca.ctxmap= ct->midword; else debugf(" (empty)"); - if (uniquematch->endsword) { + if (match->endsword) { nspaces= ct->space_spaces; debugf("_"); - ctxmap= ct->nextword; + fca.ctxmap= ct->nextword; } debugf("\n"); } else { int rx; debugf(" || UNKNOWN"); for (rx=x; rxnresults); debug_flush(); return rd->results;