PARP Research Group Universidad de Murcia


SIFT GPU wrapper functions
[Image processing]

Wrapper classes and functions for the GPU implementation of SIFT. More...

Functions

const QList< QVSiftFeaturegetSiftGPUFeatures (const QVImage< uChar > &image, nvidia_method method=CUDA_METHOD, bool computeDescriptors=FALSE, double filterWidthFactor=4.0, double orientationSampleWindow=2.0, double descriptorGridSize=3.0, int firstOctave=0, double thresholdDoG=0.02/3, double thresholdEdge=10.0, int d=3, int v=0)
 Find SIFT features in an image using user-specified parameter values, using NVIDIA CUDA.
QList< QPair< int, int > > matchSiftGPUFeatures (const QList< QVSiftFeature > img1features, const QList< QVSiftFeature > img2features, const float distmax=0.7, const float ratiomax=0.8, const bool mutual_best_match=true)
 Match SIFT features between a pair of images using using NVIDIA CUDA.

Detailed Description

Wrapper classes and functions for the GPU implementation of SIFT.

This group contains wrapper functions for the SIFT GPU implementation of Changchang Wu. You can enable the QVision compatibility with this software in the config.pri file.


Function Documentation

const QList<QVSiftFeature> getSiftGPUFeatures ( const QVImage< uChar > &  image,
nvidia_method  method = CUDA_METHOD,
bool  computeDescriptors = FALSE,
double  filterWidthFactor = 4.0,
double  orientationSampleWindow = 2.0,
double  descriptorGridSize = 3.0,
int  firstOctave = 0,
double  thresholdDoG = 0.02/3,
double  thresholdEdge = 10.0,
int  d = 3,
int  v = 0 
)

Find SIFT features in an image using user-specified parameter values, using NVIDIA CUDA.

Parameters:
image The image in which to detect features.
method 0=CUDA, 1=GLSL, 2=CG
computeDescriptors Skip descriptor computation if TRUE (not modifiable after initialization)
filterWidthFactor Factor for filter width (2*factor*sigma+1).
orientationSampleWindow Factor for orientation sample window (2*factor*sigma).
descriptorGridSize Factor for descriptor grid size (4*factor*sigma).
firstOctave First octave to start detection.
thresholdDoG DoG threshold.
thresholdEdge Edge threshold.
Returns:
Returns the list of obtained keypoints (with descriptors if sd is not TRUE).

Definition at line 44 of file qvsiftgpu.cpp.

QList< QPair<int, int> > matchSiftGPUFeatures ( const QList< QVSiftFeature img1features,
const QList< QVSiftFeature img2features,
const float  distmax = 0.7,
const float  ratiomax = 0.8,
const bool  mutual_best_match = true 
)

Match SIFT features between a pair of images using using NVIDIA CUDA.

Parameters:
img1features List of SIFT features from image 1.
img2features List of SIFT features from image 2.
distmax Maximum distance of SIFT descriptor.
ratiomax Maximum distance ratio.
mutualBestMatch Mutual best match or one way.
Returns:
Returns set of matchings. Each matching contains two index references, one for the feature in each input feature list.

Definition at line 154 of file qvsiftgpu.cpp.




QVision framework. PARP research group. Copyright © 2007, 2008, 2009, 2010, 2011.