public LatinClassifier()
{
DRandomizer.Default.init_drand(DateTime.Now.Millisecond);
charclass = new ComponentContainerIModel(IModel.MakeModel(PGet("charclass")));
junkclass = new ComponentContainerIModel(IModel.MakeModel(PGet("junkclass")));
ulclass = new ComponentContainerIModel();
PDef("junkchar", (int)'~', "junk character");
PDef("junkclass", "mlp", "junk classifier");
PDef("charclass", "mappedmlp", "character classifier");
PDef("junk", 1, "train a separate junk classifier");
PDef("ul", 0, "do upper/lower reclassification");
PDef("ulclass", "mlp", "upper/lower classifier");
junkchar = -1;
Persist(charclass, "charclass");
Persist(junkclass, "junkclass");
Persist(ulclass, "ulclass");
TryAttachCharClassifierEvent(charclass.Object);
TryAttachJunkClassifierEvent(junkclass.Object);
}
public Linerec(string classifier1 = "latin"/*none*/, string extractor1 = "scaledfe",
string segmenter1 = "DpSegmenter", int use_reject = 1)
{
transcript = "";
//line = new Bytearray();
segmentation = new Intarray();
binarized = new Bytearray();
// component choices
PDef("classifier", classifier1, "character classifier");
PDef("extractor", extractor1, "feature extractor");
PDef("segmenter", segmenter1, "line segmenter");
PDef("grouper", "SimpleGrouper", "line grouper");
// retraining
PDef("cpreload", "none", "classifier to be loaded prior to training");
// debugging
PDef("verbose", 0, "verbose output from glinerec");
// outputs
PDef("use_priors", 0, "correct the classifier output by priors");
PDef("use_reject", use_reject, "use a reject class (use posteriors only and train on junk chars)");
PDef("maxcost", 20.0, "maximum cost of a character to be added to the output");
PDef("minclass", 32, "minimum output class to be added (default=unicode space)");
PDef("minprob", 1e-9, "minimum probability for a character to appear in the output at all");
PDef("invert", 1, "invert the input line prior to char extraction");
// segmentation
PDef("maxrange", 5, "maximum number of components that are grouped together");
// sanity limits on input
PDef("minheight", 9, "minimum height of input line");
PDef("maxheight", 300, "maximum height of input line");
PDef("maxaspect", 2.0, "maximum height/width ratio of input line");
// space estimation (FIXME factor this out eventually)
PDef("space_fractile", 0.5, "fractile for space estimation");
PDef("space_multiplier", 2.0, "multipler for space estimation");
PDef("space_min", 0.2, "minimum space threshold (in xheight)");
PDef("space_max", 1.1, "maximum space threshold (in xheight)");
PDef("space_yes", 1.0, "cost of inserting a space");
PDef("space_no", 5.0, "cost of not inserting a space");
// back compability
PDef("minsize_factor", 0.0, "");
counts = new Intarray();
segmenter = new ComponentContainerISegmentLine(ComponentCreator.MakeComponent<ISegmentLine>(PGet("segmenter")));
grouper = new ComponentContainerIGrouper(ComponentCreator.MakeComponent<IGrouper>(PGet("grouper")));
classifier = new ComponentContainerIModel(IModel.MakeModel(PGet("classifier")));
TryAttachClassifierEvent(classifier.Object);
Persist(classifier, "classifier");
Persist(counts, "counts");
Persist(segmenter, "segmenter");
Persist(grouper, "grouper");
if (!classifier.IsEmpty)
{
classifier.Object.Set("junk", PGeti("use_reject"));
classifier.Object.SetExtractor(PGet("extractor"));
}
ntrained = 0;
counts_warned = false;
}
