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;
+ int (*findchar_select)(const FindCharArgs *fca,
+ const FindCharResults results[]);
};
-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"
-};
-
#define NCONTEXTS (sizeof(context_names)/sizeof(context_names[0]))
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");
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,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);
}
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);
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++;
}
-typedef struct {
- OcrReader *rd;
- int w;
- Pixcol *cols;
- int x;
- unsigned ctxmap;
-} FindCharArgs;
-
static DatabaseNode *findchar_1ctx(const FindCharArgs *fca,
DatabaseNode *start, int *matchx_r) {
DatabaseNode *current= start;
return bestmatch;
}
-static DatabaseNode *findchar_unique(const FindCharArgs *fca, int *match_rx) {
- DatabaseNode *uniquematch= 0;
- int ctxi;
-
+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]);
- DatabaseNode *match=
- findchar_1ctx(fca, &fca->rd->contexts[ctxi], match_rx);
- if (!match) continue;
+ results[ctxi].match= findchar_1ctx(fca, &fca->rd->contexts[ctxi],
+ &results[ctxi].rx);
+ if (!results[ctxi].match) continue;
- if (uniquematch && strcmp(match->str, uniquematch->str)) {
- debugf( " ambiguous");
- uniquematch= 0;
- } else {
- uniquematch= match;
- }
+ 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[]) {
+ if (fca->ctxmap != 017) return -1;
- return uniquematch;
+ 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[]) {
FindCharArgs fca;
fca.rd= rd;
+ fca.ct= ct;
fca.w= w;
fca.cols= cols;
fca.x= -1;
/* 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;
fca.x= x;
int match_rx=-1;
- DatabaseNode *match= findchar_unique(&fca, &match_rx);
+ int match_ctxi=-1;
+ DatabaseNode *match= findchar(&fca, &match_rx, &match_ctxi);
if (match) {
debugf(" || YES");
- add_result(rd, match->str, x, match_rx, fca.ctxmap);
+ 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)");
}
}
- add_result(rd, 0,-1,-1,0);
+ add_result(rd, 0,-1,-1,-1,0);
debugf("OCR finished %d glyphs\n",rd->nresults);
debug_flush();
return rd->results;