SIFT.cpp File Reference

#include <iostream>
#include <algorithm>
#include "SIFT.hpp"
#include "__SIFT.hpp"
#include "MathU.h"

Macros
#define	R_PORT   0.2126f
#define	G_PORT   0.7152f
#define	B_PORT   0.0722f
#define	MY_GAUSS   true
#define	MY_RESIZE   true

Functions
void	start_time (system_clock::time_point &startTime)
void	end_time (system_clock::time_point &startTime, system_clock::time_point &endTime, const char *step_name)
void	extractSiftFeatures (const Mat &img, vector< KeyPoint > &keypoints, Mat &descriptor, int intervals, double sigma, double contrast_thres, int curvature_thres, bool img_dbl, int descr_width, int descr_hist_bins)
void	convertGrayAndNormalize (const Mat &input, Mat &output)
void	__feats2KeyPoints (const vector< Feature > &feats, vector< KeyPoint > &keypoints)
void	__featsVec2Mat (const vector< Feature > &feats, Mat &mat)
void	filterCreation (double **GKernel, int kSize, double sigma)
void	filterCreation1D (float *GKernel, int kSize, double sigma)
void	gaussianBlur (const Mat &src, Mat &output, int kSize, double **kernel)
void	resizeHalf (const Mat &input, Mat &output, Size outSize)
Mat	__createInitImg (const Mat &img, bool img_dbl, double sigma)
void	__buildGaussPyramid (const Mat &base, vector< Mat > &gaussian_pyramid, int octaves, int intervals, double sigma)
void	__buildDogPyramid (const vector< Mat > &gaussian_pyramid, vector< Mat > &dog_pyramid, int octaves, int intervals)
void	__scaleSpaceExtrema (const vector< Mat > &dog_pyramid, vector< Feature > &feats, int octaves, int intervals, double contrast_thres, int curvature_thres)
bool	__isExtremum (const vector< Mat > &dog_pyramid, int idx, int r, int c)
bool	__interpExtremum (const vector< Mat > &dog_pyramid, Feature &feat, int idx, int r, int c, int intervals, double contrast_thres)
void	__interpStep (const vector< Mat > &dog_pyramid, int idx, int r, int c, double &xi, double &xr, double &xc)
void	_derivative ()
Mat	__derivative (const vector< Mat > &dog_pyramid, int idx, int r, int c)
Mat	__hessian (const vector< Mat > &dog_pyramid, int idx, int r, int c)
double	__interpContrast (const vector< Mat > &dog_pyramid, int idx, int r, int c, double xi, double xr, double xc)
bool	__isTooEdgeLike (const Mat &dog, int r, int c, int curvature_thres)
void	__calcFeatureScales (vector< Feature > &feats, double sigma, int intervals)
void	__adjustForImgDbl (vector< Feature > &feats)
void	__calcFeatureOris (vector< Feature > &feats, const vector< Mat > &gaussian_pyramid, int layer_per_octave)
void	__oriHist (const Mat &gaussian, vector< double > &hist, int r, int c, int rad, double sigma)
bool	__calcGradMagOri (const Mat &gaussian, int r, int c, double &mag, double &ori)
void	__smoothOriHist (vector< double > &hist)
void	__addGoodOriFeatures (queue< Feature > &feat_queue, const vector< double > &hist, double mag_thres, const Feature &feat)
void	__computeDescriptors (vector< Feature > &feats, const vector< Mat > &gaussian_pyramid, int layer_per_octave, int d, int n)
void	__descriptorHist (const Mat &gaussian, vector< double > &hist, int r, int c, double ori, double scl, int d, int n)
void	__interpHistEntry (vector< double > &hist, double rbin, double cbin, double obin, double mag, int d, int n)
void	__hist2Descriptor (const vector< double > &hist, Feature &feat, int d, int n)
void	__normalizeDescriptor (vector< double > &descriptor)

Variables
double **	gKernel
int	kSize = 9
volatile int	p_cnt = 0

Macro Definition Documentation

B_PORT

#define B_PORT   0.0722f

G_PORT

#define G_PORT   0.7152f

MY_GAUSS

#define MY_GAUSS   true

MY_RESIZE

#define MY_RESIZE   true

R_PORT

#define R_PORT   0.2126f

Function Documentation

__addGoodOriFeatures()

void __addGoodOriFeatures	(	queue< Feature > &	feat_queue,
		const vector< double > &	hist,
		double	mag_thres,
		const Feature &	feat
	)

[add features with orientation greater than mag_thres]

Parameters

feat_queue	[queue for features]
hist	[orientation histogram]
mag_thres	[magnitude threshold]
feat	[template feat]

__adjustForImgDbl()

void __adjustForImgDbl

(

vector< Feature > &

feats

)

[Halves feature coordinates and scale]

Parameters

feats

[feature vector]

__buildDogPyramid()

void __buildDogPyramid	(	const vector< Mat > &	gaussian_pyramid,
		vector< Mat > &	dog_pyramid,
		int	octaves,
		int	intervals
	)

[Build Difference of Gaussian Pyramid]

Parameters

gaussian_pyramid	[gaussian pyramid]
dog_pyramid	[returned dog pyramid, size : octaves x (intervals + 2)]
octaves	[number of octaves]
intervals	[number of intervals per octave]

__buildGaussPyramid()

void __buildGaussPyramid	(	const Mat &	base,
		vector< Mat > &	gaussian_pyramid,
		int	octaves,
		int	intervals,
		double	sigma
	)

[Build Gaussian pyramid]

Parameters

base	[base image for the pyramid]
gaussian_pyramid	[returned gaussian pyramid, size : octaves x (intervals + 3)]
octaves	[number of octaves]
intervals	[number of intervals per octave]
sigma	[sigma for gaussian smoothing]

__calcFeatureOris()

void __calcFeatureOris	(	vector< Feature > &	feats,
		const vector< Mat > &	gaussian_pyramid,
		int	layer_per_octave
	)

[Computes a canonical orientation for each image feature]

Parameters

feats	[feature vector]
gaussian_pyramid	[gaussian pyramid]
layer_per_octave	[number of layers per octave]

__calcFeatureScales()

void __calcFeatureScales	(	vector< Feature > &	feats,
		double	sigma,
		int	intervals
	)

[Calculates characteristic scale]

Parameters

feats	[feature vector]
sigma	[sigma for Gaussian smoothing]
intervals	[intervals per octave]

__calcGradMagOri()

bool __calcGradMagOri	(	const Mat &	gaussian,
		int	r,
		int	c,
		double &	mag,
		double &	ori
	)

[Calculates the gradient magnitude and orientation]

Parameters

gaussian	[Gaussian image]
r	[row]
c	[col]
mag	[returned gradient magnitude]
ori	[returned gradient orientation]

Returns

[valid pixel]

__computeDescriptors()

void __computeDescriptors	(	vector< Feature > &	feats,
		const vector< Mat > &	gaussian_pyramid,
		int	layer_per_octave,
		int	d,
		int	n
	)

[Computes feature descriptors]

Parameters

feats	[feature vector]
gaussian_pyramid	[Gaussian pyramid]
layer_per_octave	[number of layers per octave in Gaussian pyramid]
d	[width of 2d hist]
n	[number of bins for descriptor]

__createInitImg()

Mat __createInitImg	(	const Mat &	img,
		bool	img_dbl,
		double	sigma
	)

[Create the initial image for builing gaussian pyramid, optionally double the image size before smoothing]

Parameters

img	[input image]
img_dbl	[double size before smoothing?]
sigma	[sigma for gaussian smoothing]

Returns

[image returned]

__derivative()

Mat __derivative	(	const vector< Mat > &	dog_pyramid,
		int	idx,
		int	r,
		int	c
	)

[partial derivatives in x, y, interval]

Parameters

dog_pyramid	[Dog pyramid]
idx	[index of intervals]
r	[pixel's image row]
c	[pixel's image col]

Returns

[partial derivatives]

__descriptorHist()

void __descriptorHist	(	const Mat &	gaussian,
		vector< double > &	hist,
		int	r,
		int	c,
		double	ori,
		double	scl,
		int	d,
		int	n
	)

[Computes the 2D array of orientation histograms]

Parameters

gaussian	[gaussian image]
hist	[returned dxdxn hist vector]
r	[row]
c	[col]
ori	[orientation]
scl	[scl scale relative to image]
d	[width of 2d hist]
n	[number of bins for descriptor]

__feats2KeyPoints()

void __feats2KeyPoints	(	const vector< Feature > &	feats,
		vector< KeyPoint > &	keypoints
	)

[convert features to keypoints]

Parameters

feats	[features in vector]
keypoints	[returned keypoint vector]

__featsVec2Mat()

void __featsVec2Mat	(	const vector< Feature > &	feats,
		Mat &	mat
	)

[convert features from vector to Mat]

Parameters

feats	[features in vector]
mat	[returned Mat]

__hessian()

Mat __hessian	(	const vector< Mat > &	dog_pyramid,
		int	idx,
		int	r,
		int	c
	)

[3D Hessian matrix in x, y, interval]

Parameters

dog_pyramid	[Dog pyramid]
idx	[index of intervals]
r	[pixel's image row]
c	[pixel's image col]

Returns

[3D Hessian matrix]

__hist2Descriptor()

void __hist2Descriptor	(	const vector< double > &	hist,
		Feature &	feat,
		int	d,
		int	n
	)

[converts 2d hist into 1d descriptor]

Parameters

hist	[2d histogram]
feat	[feature point]
d	[width of 2d hist]
n	[number of bins for descriptor]

__interpContrast()

double __interpContrast	(	const vector< Mat > &	dog_pyramid,
		int	idx,
		int	r,
		int	c,
		double	xi,
		double	xr,
		double	xc
	)

[Calculates interpolated pixel contrast]

Parameters

dog_pyramid	[Dog pyramid]
idx	[index of intervals]
r	[pixel's image row]
c	[pixel's image col]
xi	[interpolated subpixel increment to interval]
xr	[interpolated subpixel increment to row]
xc	[interpolated subpixel increment to col]

Returns

[interpolated contrast]

__interpExtremum()

bool __interpExtremum	(	const vector< Mat > &	dog_pyramid,
		Feature &	feat,
		int	idx,
		int	r,
		int	c,
		int	intervals,
		double	contrast_thres
	)

[Interpolates a scale-space extremum's location]

Parameters

dog_pyramid	[Dog pyramid]
feat	[returned feature]
idx	[index of intervals]
r	[pixel's image row]
c	[pixel's image col]
intervals	[number of intervals per octave]
contrast_thres	[low threshold on feature contrast]

Returns

[is a feature?]

__interpHistEntry()

void __interpHistEntry	(	vector< double > &	hist,
		double	rbin,
		double	cbin,
		double	obin,
		double	mag,
		int	d,
		int	n
	)

[Interpolates an entry into the array of orientation histograms]

Parameters

hist	[2d histogram]
rbin	[row]
cbin	[col]
obin	[orientation]
mag	[size]
d	[width of 2d hist]
n	[number of bins for descriptor]

__interpStep()

void __interpStep	(	const vector< Mat > &	dog_pyramid,
		int	idx,
		int	r,
		int	c,
		double &	xi,
		double &	xr,
		double &	xc
	)

[one step of extremum interpolation]

Parameters

dog_pyramid	[Dog pyramid]
idx	[index of intervals]
r	[pixel's image row]
c	[pixel's image col]
xi	[interpolated subpixel increment to interval]
xr	[interpolated subpixel increment to row]
xc	[interpolated subpixel increment to col]

__isExtremum()

bool __isExtremum	(	const vector< Mat > &	dog_pyramid,
		int	idx,
		int	r,
		int	c
	)

[Determines whether a pixel is a scale-space extremum]

Parameters

dog_pyramid	[Dog pyramid]
idx	[index of intervals]
r	[pixel's image row]
c	[pixel's image col]

Returns

[extremum?]

__isTooEdgeLike()

bool __isTooEdgeLike	(	const Mat &	dog,
		int	r,
		int	c,
		int	curvature_thres
	)

[Is a feature too edge like?]

Parameters

dog	[Dog layer]
r	[row]
c	[col]
curvature_thres	[high threshold on ratio of principal curvatures]

Returns

[edge?]

__normalizeDescriptor()

void __normalizeDescriptor

(

vector< double > &

descriptor

)

[normalize descriptor to length 1]

Parameters

descriptor

[descriptor vector]

__oriHist()

void __oriHist	(	const Mat &	gaussian,
		vector< double > &	hist,
		int	r,
		int	c,
		int	rad,
		double	sigma
	)

[Compute gradient orientation histogram]

Parameters

gaussian	[gaussian image]
hist	[returned hist]
r	[row]
c	[col]
n	[number of bins]
rad	[radius of selected region]
sigma	[sigma for gaussian smoothing]

__scaleSpaceExtrema()

void __scaleSpaceExtrema	(	const vector< Mat > &	dog_pyramid,
		vector< Feature > &	feats,
		int	octaves,
		int	intervals,
		double	contrast_thres,
		int	curvature_thres
	)

[Detects features at extrema in DoG scale space]

Parameters

dog_pyramid	[Dog pyramid]
feats	[Detected features with scales, origentations and descriptors to be determined]
octaves	[number of octaves]
intervals	[number of intervals per octave]
contrast_thres	[low threshold on feature contrast]
curvature_thres	[high threshold on feature ratio of principal curvatures]

__smoothOriHist()

void __smoothOriHist

(

vector< double > &

hist

)

[Gaussian smooths an orientation histogram.]

Parameters

hist	[orientation histogram]

_derivative()

void _derivative

(

)

convertGrayAndNormalize()

void convertGrayAndNormalize	(	const Mat &	input,
		Mat &	output
	)

Change 3Channel RGB image to normalized gray image.

Parameters

input	3Channel RGB input image
output	normalized grayscale output image with pixel values in 0.0f~1.0f

end_time()

void end_time	(	system_clock::time_point &	startTime,
		system_clock::time_point &	endTime,
		const char *	step_name
	)

extractSiftFeatures()

void extractSiftFeatures	(	const Mat &	img,
		vector< KeyPoint > &	keypoints,
		Mat &	descriptor,
		int	intervals = SIFT_INTERVALS,
		double	sigma = SIFT_SIGMA,
		double	contrast_thres = SIFT_CONTRAST_THRES,
		int	curvature_thres = SIFT_CURVATURE_THRES,
		bool	img_dbl = SIFT_IMG_DBL,
		int	descr_width = SIFT_DESCR_WIDTH,
		int	descr_hist_bins = SIFT_DESCR_HIST_BINS
	)

[Detect & extract sift features for an image]

Parameters

img	[source image for extract sift features : grayscale image with pixel values in 0.0f~1.0f]
keypoints	[vector for store sift feature point]
descriptor	[Mat for store sift descriptor, each row for a keypoint]
intervals	[the number of sampled intervals per octave]
sigma	[sigma for initial gaussian smoothing]
contrast_thres	[threshold on keypoint contrast]
curvature_thres	[threshold on keypoint ratio of principle curvatures]
img_dbl	[double image size before pyramid construction?]
descr_width	[width of descriptor histogram array]
descr_hist_bins	[number of bins per histogram in descriptor array]

filterCreation()

void filterCreation	(	double **	GKernel,
		int	kSize,
		double	sigma
	)

Create 2D gaussian filter

Parameters

GKernel	output 2D gaussian filter
kSize	gaussian kernel size
sigma	gaussian sigma

filterCreation1D()

void filterCreation1D	(	float *	GKernel,
		int	kSize,
		double	sigma
	)

Create 1D gaussian filter

Parameters

GKernel	output 1D gaussian filter
kSize	gaussian kernel size
sigma	gaussian sigma

gaussianBlur()

void gaussianBlur	(	const Mat &	src,
		Mat &	output,
		int	kSize,
		double **	kernel
	)

Blurs an image using a Gaussian filter using padding
Assume that dst already copy the data of src

Parameters

src	input image; Normalized grayscale input image
output	output image of the same size and type as src.
kSize	width of the input image
kernel	Gaussian 2D kernel

resizeHalf()

void resizeHalf	(	const Mat &	input,
		Mat &	output,
		Size	outSize
	)

Resize image using Nearest neighbor sampling. Reduce width and height in half. (Native)

Parameters

input	normalized grayscale input image
output	resized output image
outSize	output image size

start_time()

void start_time

(

system_clock::time_point &

startTime

)

Variable Documentation

gKernel

double** gKernel

kSize

int kSize = 9

p_cnt

volatile int p_cnt = 0