mftraining.cpp

来自「一个google的OCR源码」· C++ 代码 · 共 1,342 行 · 第 1/3 页

	fprintf(File, "%s\n", MergeClass->Label);
	fprintf(File, "%d\n", NumProtosIn(MergeClass->Class));
	for(i=0; i < NumProtosIn(MergeClass->Class); i++)
	{
		Proto = ProtoIn(MergeClass->Class,i);
		fprintf(File, "\t%8.4f %8.4f %8.4f %8.4f ", ProtoX(Proto), ProtoY(Proto),
			ProtoLength(Proto), ProtoAngle(Proto));
		Values[0] = ProtoX(Proto);
		Values[1] = ProtoY(Proto);
		Values[2] = ProtoAngle(Proto);
		Normalize(Values);
		fprintf(File, "%8.4f %8.4f %8.4f\n", Values[0], Values[1], Values[2]);
	}
} // WriteProtos

/*----------------------------------------------------------------------------*/
void WriteConfigs(
	FILE* File,
	CLASS_TYPE Class)
{
	BIT_VECTOR Config;
	int i, j, WordsPerConfig;

	WordsPerConfig = WordsInVectorOfSize(NumProtosIn(Class));
	fprintf(File, "%d %d\n", NumConfigsIn(Class),WordsPerConfig);
	for(i=0; i < NumConfigsIn(Class); i++)
	{
		Config = ConfigIn(Class,i);
		for(j=0; j < WordsPerConfig; j++)
			fprintf(File, "%08x ", Config[j]);
		fprintf(File, "\n");
	}
	fprintf(File, "\n");
} // WriteConfigs

/*---------------------------------------------------------------------------*/
void FreeTrainingSamples (
     LIST	CharList)

/*
**	Parameters:
**		FontList	list of all fonts in document
**	Globals: none
**	Operation:
**		This routine deallocates all of the space allocated to
**		the specified list of training samples.
**	Return: none
**	Exceptions: none
**	History: Fri Aug 18 17:44:27 1989, DSJ, Created.
*/

{
	LABELEDLIST	CharSample;
	FEATURE_SET	FeatureSet;
	LIST		FeatureList;


// 	printf ("FreeTrainingSamples...\n");
	iterate (CharList) 		/* iterate thru all of the fonts */
	{
		CharSample = (LABELEDLIST) first_node (CharList);
		FeatureList = CharSample->List;
		iterate (FeatureList)	/* iterate thru all of the classes */
		{
			FeatureSet = (FEATURE_SET) first_node (FeatureList);
			FreeFeatureSet (FeatureSet);
		}
		FreeLabeledList (CharSample);
	}
	destroy (CharList);

}	/* FreeTrainingSamples */

/*-----------------------------------------------------------------------------*/
void FreeLabeledClassList (
     LIST	ClassList)

/*
**	Parameters:
**		FontList	list of all fonts in document
**	Globals: none
**	Operation:
**		This routine deallocates all of the space allocated to
**		the specified list of training samples.
**	Return: none
**	Exceptions: none
**	History: Fri Aug 18 17:44:27 1989, DSJ, Created.
*/

{
	MERGE_CLASS	MergeClass;

	iterate (ClassList) 		/* iterate thru all of the fonts */
	{
		MergeClass = (MERGE_CLASS) first_node (ClassList);
		free (MergeClass->Label);
		FreeClass(MergeClass->Class);
		free (MergeClass);
	}
	destroy (ClassList);

}	/* FreeLabeledClassList */

/*---------------------------------------------------------------------------*/
void FreeLabeledList (
     LABELEDLIST	LabeledList)

/*
**	Parameters:
**		LabeledList	labeled list to be freed
**	Globals: none
**	Operation:
**		This routine deallocates all of the memory consumed by
**		a labeled list.  It does not free any memory which may be
**		consumed by the items in the list.
**	Return: none
**	Exceptions: none
**	History: Fri Aug 18 17:52:45 1989, DSJ, Created.
*/

{
	destroy (LabeledList->List);
	free (LabeledList->Label);
	free (LabeledList);

}	/* FreeLabeledList */

/*---------------------------------------------------------------------------*/
CLUSTERER *SetUpForClustering(
     LABELEDLIST	CharSample)

/*
**	Parameters:
**		CharSample: LABELEDLIST that holds all the feature information for a
**		given character.
**	Globals:
**		None
**	Operation:
**		This routine reads samples from a LABELEDLIST and enters
**		those samples into a clusterer data structure.  This
**		data structure is then returned to the caller.
**	Return:
**		Pointer to new clusterer data structure.
**	Exceptions:
**		None
**	History:
**		8/16/89, DSJ, Created.
*/

{
	uinT16	N;
	int		i, j;
	FLOAT32	*Sample = NULL;
	CLUSTERER	*Clusterer;
	inT32		CharID;
	LIST FeatureList = NULL;
	FEATURE_SET FeatureSet = NULL;
	FEATURE_DESC FeatureDesc = NULL;
//	PARAM_DESC* ParamDesc;

	FeatureDesc = DefinitionOf(ShortNameToFeatureType(PROGRAM_FEATURE_TYPE));
	N = FeatureDesc->NumParams;
//	ParamDesc = ConvertToPARAMDESC(FeatureDesc->ParamDesc, N);
	Clusterer = MakeClusterer(N,FeatureDesc->ParamDesc);
//	free(ParamDesc);

	FeatureList = CharSample->List;
	CharID = 0;
	iterate(FeatureList)
	{
		if (CharID >= MaxNumSamples) break;

		FeatureSet = (FEATURE_SET) first_node (FeatureList);
		for (i=0; i < FeatureSet->MaxNumFeatures; i++)
		{
			if (Sample == NULL)
				Sample = (FLOAT32 *)Emalloc(N * sizeof(FLOAT32));
			for (j=0; j < N; j++)
				if (RoundingAccuracy != 0.0f)
					Sample[j] = round(FeatureSet->Features[i]->Params[j], RoundingAccuracy);
				else
					Sample[j] = FeatureSet->Features[i]->Params[j];
				MakeSample (Clusterer, Sample, CharID);
		}
		CharID++;
	}
	if ( Sample != NULL ) free( Sample );
	return( Clusterer );

}	/* SetUpForClustering */

/*------------------------------------------------------------------------*/
void MergeInsignificantProtos(LIST ProtoList, const char* label,
                              CLUSTERER	*Clusterer, CLUSTERCONFIG *Config) {
  PROTOTYPE	*Prototype;
  bool debug = strcmp(test_ch, label) == 0;

  LIST pProtoList = ProtoList;
  iterate(pProtoList) {
    Prototype = (PROTOTYPE *) first_node (pProtoList);
    if (Prototype->Significant || Prototype->Merged)
      continue;
    FLOAT32 best_dist = 0.125;
    PROTOTYPE* best_match = NULL;
    // Find the nearest alive prototype.
    LIST list_it = ProtoList;
    iterate(list_it) {
      PROTOTYPE* test_p = (PROTOTYPE *) first_node (list_it);
      if (test_p != Prototype && !test_p->Merged) {
        FLOAT32 dist = ComputeDistance(Clusterer->SampleSize,
                                       Clusterer->ParamDesc,
                                       Prototype->Mean, test_p->Mean);
        if (dist < best_dist) {
          best_match = test_p;
          best_dist = dist;
        }
      }
    }
    if (best_match != NULL && !best_match->Significant) {
      if (debug)
         tprintf("Merging red clusters (%d+%d) at %g,%g and %g,%g\n",
                 best_match->NumSamples, Prototype->NumSamples,
                 best_match->Mean[0], best_match->Mean[1],
                 Prototype->Mean[0], Prototype->Mean[1]);
      best_match->NumSamples = MergeClusters(Clusterer->SampleSize,
                                             Clusterer->ParamDesc,
                                             best_match->NumSamples,
                                             Prototype->NumSamples,
                                             best_match->Mean,
                                             best_match->Mean, Prototype->Mean);
      Prototype->NumSamples = 0;
      Prototype->Merged = 1;
    } else if (best_match != NULL) {
      if (debug)
        tprintf("Red proto at %g,%g matched a green one at %g,%g\n",
                Prototype->Mean[0], Prototype->Mean[1],
                best_match->Mean[0], best_match->Mean[1]);
      Prototype->Merged = 1;
    }
  }
  // Mark significant those that now have enough samples.
  int min_samples = (inT32) (Config->MinSamples * Clusterer->NumChar);
  pProtoList = ProtoList;
  iterate(pProtoList) {
    Prototype = (PROTOTYPE *) first_node (pProtoList);
    // Process insignificant protos that do not match a green one
    if (!Prototype->Significant && Prototype->NumSamples >= min_samples &&
        !Prototype->Merged) {
      if (debug)
        tprintf("Red proto at %g,%g becoming green\n",
                Prototype->Mean[0], Prototype->Mean[1]);
      Prototype->Significant = true;
    }
  }
}	/* MergeInsignificantProtos */

/*------------------------------------------------------------------------*/
LIST RemoveInsignificantProtos(
	LIST ProtoList,
	BOOL8 KeepSigProtos,
	BOOL8 KeepInsigProtos,
	int N)

{
	LIST NewProtoList = NIL;
	LIST pProtoList;
	PROTOTYPE* Proto;
	PROTOTYPE* NewProto;
	int i;

	pProtoList = ProtoList;
	iterate(pProtoList)
	{
		Proto = (PROTOTYPE *) first_node (pProtoList);
		if ((Proto->Significant && KeepSigProtos) ||
			(!Proto->Significant && KeepInsigProtos))
		{
			NewProto = (PROTOTYPE *)Emalloc(sizeof(PROTOTYPE));

			NewProto->Mean = (FLOAT32 *)Emalloc(N * sizeof(FLOAT32));
			NewProto->Significant = Proto->Significant;
			NewProto->Style = Proto->Style;
			NewProto->NumSamples = Proto->NumSamples;
			NewProto->Cluster = NULL;
			NewProto->Distrib = NULL;

			for (i=0; i < N; i++)
				NewProto->Mean[i] = Proto->Mean[i];
			if (Proto->Variance.Elliptical != NULL)
			{
				NewProto->Variance.Elliptical = (FLOAT32 *)Emalloc(N * sizeof(FLOAT32));
				for (i=0; i < N; i++)
					NewProto->Variance.Elliptical[i] = Proto->Variance.Elliptical[i];
			}
			else
				NewProto->Variance.Elliptical = NULL;
			//---------------------------------------------
			if (Proto->Magnitude.Elliptical != NULL)
			{
				NewProto->Magnitude.Elliptical = (FLOAT32 *)Emalloc(N * sizeof(FLOAT32));
				for (i=0; i < N; i++)
					NewProto->Magnitude.Elliptical[i] = Proto->Magnitude.Elliptical[i];
			}
			else
				NewProto->Magnitude.Elliptical = NULL;
			//------------------------------------------------
			if (Proto->Weight.Elliptical != NULL)
			{
				NewProto->Weight.Elliptical = (FLOAT32 *)Emalloc(N * sizeof(FLOAT32));
				for (i=0; i < N; i++)
					NewProto->Weight.Elliptical[i] = Proto->Weight.Elliptical[i];
			}
			else
				NewProto->Weight.Elliptical = NULL;

			NewProto->TotalMagnitude = Proto->TotalMagnitude;
			NewProto->LogMagnitude = Proto->LogMagnitude;
			NewProtoList = push_last(NewProtoList, NewProto);
		}
	}
	//FreeProtoList (ProtoList);
	return (NewProtoList);
}	/* RemoveInsignificantProtos */
/*-----------------------------------------------------------------------------*/
void CleanUpUnusedData(
	LIST ProtoList)
{
	PROTOTYPE* Prototype;

	iterate(ProtoList)
	{
		Prototype = (PROTOTYPE *) first_node (ProtoList);
		if(Prototype->Variance.Elliptical != NULL)
		{
			memfree(Prototype->Variance.Elliptical);
			Prototype->Variance.Elliptical = NULL;
		}
		if(Prototype->Magnitude.Elliptical != NULL)
		{
			memfree(Prototype->Magnitude.Elliptical);
			Prototype->Magnitude.Elliptical = NULL;
		}
		if(Prototype->Weight.Elliptical != NULL)
		{
			memfree(Prototype->Weight.Elliptical);
			Prototype->Weight.Elliptical = NULL;
		}
	}
}

/*--------------------------------------------------------------------------*/
void Normalize (
   float  *Values)
{
	register float Slope;
	register float Intercept;
	register float Normalizer;

	Slope      = tan (Values [2] * 2 * PI);
	Intercept  = Values [1] - Slope * Values [0];
	Normalizer = 1 / sqrt (Slope * Slope + 1.0);

	Values [0] = Slope * Normalizer;
	Values [1] = - Normalizer;
	Values [2] = Intercept * Normalizer;
} // Normalize

/** SetUpForFloat2Int **************************************************/
void SetUpForFloat2Int(
	LIST LabeledClassList)
{
	MERGE_CLASS	MergeClass;
	CLASS_TYPE		Class;
	int				NumProtos;
	int				NumConfigs;
	int				NumWords;
	int				i, j;
	float			Values[3];
	PROTO			NewProto;
	PROTO			OldProto;
	BIT_VECTOR		NewConfig;
	BIT_VECTOR		OldConfig;

// 	printf("Float2Int ...\n");

	iterate(LabeledClassList)
	{
		MergeClass = (MERGE_CLASS) first_node (LabeledClassList);
		Class = &TrainingData[unicharset_mftraining.unichar_to_id(
                                          MergeClass->Label)];
		NumProtos = NumProtosIn(MergeClass->Class);
		NumConfigs = NumConfigsIn(MergeClass->Class);

		NumProtosIn(Class) = NumProtos;
		Class->MaxNumProtos = NumProtos;
		Class->Prototypes = (PROTO) Emalloc (sizeof(PROTO_STRUCT) * NumProtos);
		for(i=0; i < NumProtos; i++)
		{
			NewProto = ProtoIn(Class, i);
			OldProto = ProtoIn(MergeClass->Class, i);
			Values[0] = ProtoX(OldProto);
			Values[1] = ProtoY(OldProto);
			Values[2] = ProtoAngle(OldProto);
			Normalize(Values);
			ProtoX(NewProto) = ProtoX(OldProto);
			ProtoY(NewProto) = ProtoY(OldProto);
			ProtoLength(NewProto) = ProtoLength(OldProto);
			ProtoAngle(NewProto) = ProtoAngle(OldProto);
			CoefficientA(NewProto) = Values[0];
			CoefficientB(NewProto) = Values[1];
			CoefficientC(NewProto) = Values[2];
		}

		NumConfigsIn(Class) = NumConfigs;
		Class->MaxNumConfigs = NumConfigs;
		Class->Configurations = (BIT_VECTOR*) Emalloc (sizeof(BIT_VECTOR) * NumConfigs);
		NumWords = WordsInVectorOfSize(NumProtos);
		for(i=0; i < NumConfigs; i++)
		{
			NewConfig = NewBitVector(NumProtos);
			OldConfig = ConfigIn(MergeClass->Class, i);
			for(j=0; j < NumWords; j++)
				NewConfig[j] = OldConfig[j];
			ConfigIn(Class, i) = NewConfig;
		}
	}
} // SetUpForFloat2Int

/*--------------------------------------------------------------------------*/
void WritePFFMTable(INT_TEMPLATES Templates, const char* filename) {
  FILE* fp = Efopen(filename, "wb");
  /* then write out each class */
  for (int i = 0; i < NumClassesIn (Templates); i++) {
    int MaxLength = 0;
    INT_CLASS Class = ClassForIndex (Templates, i);
    for (int ConfigId = 0; ConfigId < NumIntConfigsIn (Class); ConfigId++) {
      if (LengthForConfigId (Class, ConfigId) > MaxLength)
        MaxLength = LengthForConfigId (Class, ConfigId);
    }
    fprintf(fp, "%s %d\n", unicharset_mftraining.id_to_unichar(
                ClassIdForIndex(Templates, i)), MaxLength);
  }
  fclose(fp);
} // WritePFFMTable