// Removes any result that has all its unichars covered by a better choice,
// regardless of font.
void ShapeClassifier::FilterDuplicateUnichars(
GenericVector<ShapeRating>* results) const {
GenericVector<ShapeRating> filtered_results;
// Copy results to filtered results and knock out duplicate unichars.
const ShapeTable* shapes = GetShapeTable();
for (int r = 0; r < results->size(); ++r) {
if (r > 0) {
const Shape& shape_r = shapes->GetShape((*results)[r].shape_id);
int c;
for (c = 0; c < shape_r.size(); ++c) {
int unichar_id = shape_r[c].unichar_id;
int s;
for (s = 0; s < r; ++s) {
const Shape& shape_s = shapes->GetShape((*results)[s].shape_id);
if (shape_s.ContainsUnichar(unichar_id))
break; // We found unichar_id.
}
if (s == r)
break; // We didn't find unichar_id.
}
if (c == shape_r.size())
continue; // We found all the unichar ids in previous answers.
}
filtered_results.push_back((*results)[r]);
}
*results = filtered_results;
}
void ShapeClassifier::PrintResults(
const char* context, const GenericVector<ShapeRating>& results) const {
tprintf("%s\n", context);
for (int i = 0; i < results.size(); ++i) {
tprintf("%g:", results[i].rating);
if (results[i].joined)
tprintf("[J]");
if (results[i].broken)
tprintf("[B]");
tprintf(" %s\n", GetShapeTable()->DebugStr(results[i].shape_id).string());
}
}
// Returns the shape that contains unichar_id that has the best result.
// If result is not NULL, it is set with the shape_id and rating.
// Does not need to be overridden if ClassifySample respects the keep_this
// rule.
int ShapeClassifier::BestShapeForUnichar(const TrainingSample& sample,
Pix* page_pix, UNICHAR_ID unichar_id,
ShapeRating* result) {
GenericVector<ShapeRating> results;
const ShapeTable* shapes = GetShapeTable();
int num_results = ClassifySample(sample, page_pix, 0, unichar_id, &results);
for (int r = 0; r < num_results; ++r) {
if (shapes->GetShape(results[r].shape_id).ContainsUnichar(unichar_id)) {
if (result != NULL)
*result = results[r];
return results[r].shape_id;
}
}
return -1;
}
// Classifies the given [training] sample, writing to results.
// See shapeclassifier.h for a full description.
// Default implementation calls the ShapeRating version.
int ShapeClassifier::UnicharClassifySample(
const TrainingSample& sample, Pix* page_pix, int debug,
UNICHAR_ID keep_this, GenericVector<UnicharRating>* results) {
results->truncate(0);
GenericVector<ShapeRating> shape_results;
int num_shape_results = ClassifySample(sample, page_pix, debug, keep_this,
&shape_results);
const ShapeTable* shapes = GetShapeTable();
GenericVector<int> unichar_map;
unichar_map.init_to_size(shapes->unicharset().size(), -1);
for (int r = 0; r < num_shape_results; ++r) {
shapes->AddShapeToResults(shape_results[r], &unichar_map, results);
}
return results->size();
}
// Displays classification as the given shape_id. Creates as many windows
// as it feels fit, using index as a guide for placement. Adds any created
// windows to the windows output and returns a new index that may be used
// by any subsequent classifiers. Caller waits for the user to view and
// then destroys the windows by clearing the vector.
int TessClassifier::DisplayClassifyAs(
const TrainingSample& sample, Pix* page_pix, int unichar_id, int index,
PointerVector<ScrollView>* windows) {
int shape_id = unichar_id;
if (GetShapeTable() != NULL)
shape_id = BestShapeForUnichar(sample, page_pix, unichar_id, NULL);
if (shape_id < 0) return index;
if (UnusedClassIdIn(classify_->PreTrainedTemplates, shape_id)) {
tprintf("No built-in templates for class/shape %d\n", shape_id);
return index;
}
classify_->ShowBestMatchFor(shape_id, sample.features(),
sample.num_features());
return index;
}
// Provides access to the UNICHARSET that this classifier works with.
// Only needs to be overridden if GetShapeTable() can return NULL.
const UNICHARSET& ShapeClassifier::GetUnicharset() const {
return GetShapeTable()->unicharset();
}
