drwnPatchMatchGraphLearner Class Reference

Learns a PatchMatchGraph by iteratively performing search moves over the space of matches. More...

Public Member Functions
	drwnPatchMatchGraphLearner (drwnPatchMatchGraph &graph)
	construct an object to optimize the given patch match graph
void	initialize ()
	randomly initialize matches for all active images (keeps existing matches unless an improvement is found) More...
virtual void	initialize (unsigned imgIndx)
	randomly initialize matches for a given image
void	rescore ()
	rescore matches if scoring function or features have changed (e.g., greyscale to rgb)
void	update ()
	perform one edge update iteration (including enrichment and exhuastive) on all active images
virtual void	update (unsigned imgIndx)
	perform one edge update iteration (excluding enrichment) on a single image

Static Public Attributes
static double	SEARCH_DECAY_RATE = 0.5
	decay rate during search in range [0, 1)
static int	FORWARD_ENRICHMENT_K = 3
	forward enrichment search depth
static bool	DO_INVERSE_ENRICHMENT = true
	perform inverse enrichment
static bool	DO_LOCAL_SEARCH = true
	run neighbourhood search on dirty pixels
static bool	DO_EXHAUSTIVE = false
	run exhustive search on random pixel
static drwnPatchMatchTransform	ALLOWED_TRANSFORMATIONS = DRWN_PM_TRANSFORM_NONE
	allowable patch transformations
static double	TOP_VAR_PATCHES = 1.0
	initialize all patches below this with only one match

Protected Member Functions
void	cacheImageFeatures ()
void	cacheImageFeatures (unsigned imgIndx)
	caches features for given image pyramid
virtual void	cacheImageFeatures (unsigned imgIndx, const cv::Mat &baseImg)
	caches features for given image pyramid using given base image
int	appendCIELabFeatures (const cv::Mat &img, cv::Mat &features, int nChannel=0) const
	helper functions for feature computation More...
int	appendVerticalFeatures (const cv::Mat &img, cv::Mat &features, int nChannel=0) const
int	appendEdgeFeatures (const cv::Mat &img, cv::Mat &features, int nChannel=0) const
int	appendTextonFilterFeatures (const cv::Mat &img, cv::Mat &features, int nChannel=0) const
bool	propagate (const drwnPatchMatchNode &u, bool bDirection)
	propagate good matches from patch u to its neighbours (returns true if a better match was found)
bool	search (const drwnPatchMatchNode &u)
	random exponentially decaying neighbourhood search for patch pixIndx in image imgIndx (returns true if a better match was found)
bool	local (const drwnPatchMatchNode &u)
	local neighbourhood search around patch pixIndx in image imgIndx (returns true if a better match was found)
virtual bool	enrichment ()
	enrichment: inverse (from target to source) and forward (from source to target's target) (returns true if a better match was found)
bool	exhaustive (const drwnPatchMatchNode &u)
	exhaustive search for best match across entire graph — use sparingly (returns true if a better match was found)
virtual float	scoreMatch (const drwnPatchMatchNode &u, const drwnPatchMatchNode &v, const drwnPatchMatchTransform &xform, float maxValue=DRWN_FLT_MAX) const
	match scoring function (maxValue allows for early termination)
cv::Point	mapPatch (const cv::Point &p, int imgIndx, int srcScale, int dstScale) const
	map a patch from one image pyramid level to another

Protected Attributes
drwnPatchMatchGraph &	_graph
	the graph that is being learned
vector< vector< cv::Mat > >	_features
	image features indexed by (image, pyramid level)

Detailed Description

Learns a PatchMatchGraph by iteratively performing search moves over the space of matches.

See also

PatchMatchGraph Project

Member Function Documentation

appendCIELabFeatures()

int drwnPatchMatchGraphLearner::appendCIELabFeatures ( const cv::Mat &  img, cv::Mat &  features, int  nChannel = 0  ) const

protected

helper functions for feature computation

Todo:

move these elsewhere

cacheImageFeatures()

void drwnPatchMatchGraphLearner::cacheImageFeatures ( )

protected

Todo:

allow a pixel feature generator to be provided caches features for all image pyramids

initialize()

void drwnPatchMatchGraphLearner::initialize

(

)

randomly initialize matches for all active images (keeps existing matches unless an improvement is found)

Todo:

cache image features here instead of in constructor?

---

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

• drwnPatchMatch.h
• drwnPatchMatch.cpp