Tesseract  3.02
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
pageres.h File Reference
#include "blobs.h"
#include "boxword.h"
#include "elst.h"
#include "genericvector.h"
#include "normalis.h"
#include "ocrblock.h"
#include "ocrrow.h"
#include "params_training_featdef.h"
#include "ratngs.h"
#include "rejctmap.h"
#include "seam.h"
#include "werd.h"

Go to the source code of this file.

Classes

struct  BlamerBundle
class  PAGE_RES
class  BLOCK_RES
class  ROW_RES
class  WERD_RES
class  PAGE_RES_IT

Namespaces

namespace  tesseract

Enumerations

enum  IncorrectResultReason {
  IRR_CORRECT, IRR_CLASSIFIER, IRR_CHOPPER, IRR_CLASS_LM_TRADEOFF,
  IRR_PAGE_LAYOUT, IRR_SEGSEARCH_HEUR, IRR_SEGSEARCH_PP, IRR_CLASS_OLD_LM_TRADEOFF,
  IRR_ADAPTION, IRR_NO_TRUTH_SPLIT, IRR_NO_TRUTH, IRR_UNKNOWN,
  IRR_NUM_REASONS
}
enum  CRUNCH_MODE { CR_NONE, CR_KEEP_SPACE, CR_LOOSE_SPACE, CR_DELETE }

Functions

 ELISTIZEH (BLOCK_RES) CLISTIZEH(BLOCK_RES) class ROW_RES

Enumeration Type Documentation

Enumerator:
CR_NONE 
CR_KEEP_SPACE 
CR_LOOSE_SPACE 
CR_DELETE 

Definition at line 304 of file pageres.h.

Enumerator:
IRR_CORRECT 
IRR_CLASSIFIER 
IRR_CHOPPER 
IRR_CLASS_LM_TRADEOFF 
IRR_PAGE_LAYOUT 
IRR_SEGSEARCH_HEUR 
IRR_SEGSEARCH_PP 
IRR_CLASS_OLD_LM_TRADEOFF 
IRR_ADAPTION 
IRR_NO_TRUTH_SPLIT 
IRR_NO_TRUTH 
IRR_UNKNOWN 
IRR_NUM_REASONS 

Definition at line 45 of file pageres.h.

{
// The text recorded in best choice == truth text
// Either: Top choice is incorrect and is a dictionary word (language model
// is unlikely to help correct such errors, so blame the classifier).
// Or: the correct unichar was not included in shortlist produced by the
// classifier at all.
// Chopper have not found one or more splits that correspond to the correct
// character bounding boxes recorded in BlamerBundle::truth_word.
// Classifier did include correct unichars for each blob in the correct
// segmentation, however its rating could have been too bad to allow the
// language model to pull out the correct choice. On the other hand the
// strength of the language model might have been too weak to favor the
// correct answer, this we call this case a classifier-language model
// tradeoff error.
// Page layout failed to produce the correct bounding box. Blame page layout
// if the truth was not found for the word, which implies that the bounding
// box of the word was incorrect (no truth word had a similar bounding box).
// SegSearch heuristic prevented one or more blobs from the correct
// segmentation state to be classified (e.g. the blob was too wide).
// The correct segmentaiton state was not explored because of poor SegSearch
// pain point prioritization. We blame SegSearch pain point prioritization
// if the best rating of a choice constructed from correct segmentation is
// better than that of the best choice (i.e. if we got to explore the correct
// segmentation state, language model would have picked the correct choice).
// Same as IRR_CLASS_LM_TRADEOFF, but used when we only run chopper on a word,
// and thus use the old language model (permuters).
// TODO(antonova): integrate the new language mode with chopper
// If there is an incorrect adaptive template match with a better score than
// a correct one (either pre-trained or adapted), mark this as adaption error.
// split_and_recog_word() failed to find a suitable split in truth.
// Truth is not available for this word (e.g. when words in corrected content
// file are turned into ~~~~ because an appropriate alignment was not found.
// The text recorded in best choice != truth text, but none of the above
// reasons are set.
};

Function Documentation

ELISTIZEH ( BLOCK_RES  )