vt::Database Class Reference

Class for efficiently matching a bag-of-words representation of a document (image) against a database of known documents. More...

#include <database.h>

List of all members.

Classes
struct	WordFrequency
Public Member Functions
void	computeTfIdfWeights (float default_weight=1.0f)
	Compute the TF-IDF weights of all the words. To be called after inserting a corpus of training examples into the database.
	Database (uint32_t num_words=0)
	Constructor.
void	find (const std::vector< Word > &document, size_t N, std::vector< Match > &matches) const
	Find the top N matches in the database for the query document.
DocId	findAndInsert (const std::vector< Word > &document, size_t N, std::vector< Match > &matches)
	Find the top N matches, then insert the query document.
DocId	insert (const std::vector< Word > &document)
	Insert a new document.
void	loadWeights (const std::string &file)
	Load the vocabulary word weights from a file.
void	saveWeights (const std::string &file) const
	Save the vocabulary word weights to a file.
Private Types
typedef std::map< Word, float >	DocumentVector
typedef std::vector < WordFrequency >	InvertedFile
Private Member Functions
void	computeVector (const std::vector< Word > &document, DocumentVector &v) const
Static Private Member Functions
static void	normalize (DocumentVector &v)
static float	sparseDistance (const DocumentVector &v1, const DocumentVector &v2)
Private Attributes
std::vector< DocumentVector >	database_vectors_
std::vector< InvertedFile >	word_files_
std::vector< float >	word_weights_

---

Detailed Description

Class for efficiently matching a bag-of-words representation of a document (image) against a database of known documents.

Definition at line 40 of file database.h.

---

Member Typedef Documentation

typedef std::map<Word, float> vt::Database::DocumentVector [private]

Todo:

Use sorted vector?

Definition at line 110 of file database.h.

typedef std::vector<WordFrequency> vt::Database::InvertedFile [private]

Definition at line 106 of file database.h.

---

Constructor & Destructor Documentation

vt::Database::Database

(

uint32_t

num_words = 0

)

Constructor.

If computing weights for a new vocabulary, num_words should be the size of the vocabulary. If calling loadWeights(), it can be left zero.

Definition at line 11 of file database.cpp.

---

Member Function Documentation

void vt::Database::computeTfIdfWeights

(

float

default_weight = 1.0f

)

Compute the TF-IDF weights of all the words. To be called after inserting a corpus of training examples into the database.

Parameters:

default_weight

The default weight of a word that appears in none of the training documents.

Definition at line 67 of file database.cpp.

void vt::Database::computeVector	(	const std::vector< Word > &	document,
		DocumentVector &	v
	)		const [private]

Definition at line 106 of file database.cpp.

void vt::Database::find	(	const std::vector< Word > &	document,
		size_t	N,
		std::vector< Match > &	matches
	)		const

Find the top N matches in the database for the query document.

Parameters:

	document	The query document, a set of quantized words.
	N	The number of matches to return.
[out]	matches	IDs and scores for the top N matching database documents.

Todo:

Try only computing distances against documents sharing at least one word

Definition at line 39 of file database.cpp.

DocId vt::Database::findAndInsert	(	const std::vector< Word > &	document,
		size_t	N,
		std::vector< Match > &	matches
	)

Find the top N matches, then insert the query document.

This is equivalent to calling find() followed by insert(), but may be more efficient.

Parameters:

	document	The document to match then insert, a set of quantized words.
	N	The number of matches to return.
[out]	matches	IDs and scores for the top N matching database documents.

Todo:

Can this be accelerated? Could iterate over words only once?

Definition at line 60 of file database.cpp.

DocId vt::Database::insert

(

const std::vector< Word > &

document

)

Insert a new document.

Parameters:

document

The set of quantized words in a document/image.

Returns:

An ID representing the inserted document.

Todo:

Evaluate whether sorting words makes much difference in speed

Definition at line 17 of file database.cpp.

void vt::Database::loadWeights

(

const std::string &

file

)

Load the vocabulary word weights from a file.

Definition at line 88 of file database.cpp.

void vt::Database::normalize

(

DocumentVector &

v

)

[static, private]

Definition at line 115 of file database.cpp.

void vt::Database::saveWeights

(

const std::string &

file

)

const

Save the vocabulary word weights to a file.

Definition at line 80 of file database.cpp.

float vt::Database::sparseDistance	(	const DocumentVector &	v1,
		const DocumentVector &	v2
	)		[static, private]

Definition at line 125 of file database.cpp.

---

Member Data Documentation

std::vector<DocumentVector> vt::Database::database_vectors_ [private]

Definition at line 114 of file database.h.

std::vector<InvertedFile> vt::Database::word_files_ [private]

Definition at line 112 of file database.h.

std::vector<float> vt::Database::word_weights_ [private]

Definition at line 113 of file database.h.

---

The documentation for this class was generated from the following files:

• database.h
• database.cpp