org.openimaj.image.feature.dense.gradient.dsift

Class ApproximateDenseSIFT

java.lang.Object

org.openimaj.image.feature.dense.gradient.dsift.AbstractDenseSIFT<FImage>

org.openimaj.image.feature.dense.gradient.dsift.DenseSIFT

org.openimaj.image.feature.dense.gradient.dsift.ApproximateDenseSIFT

All Implemented Interfaces:

Cloneable, ImageAnalyser<FImage>

public class ApproximateDenseSIFT
extends DenseSIFT

Implementation of an approximate dense SIFT feature extractor. Extracts approximate upright SIFT features at a single scale on a grid. Implementation is approximate because instead of using an exact Gaussian weighting, samples are weighted using a flat windowing function for speed, and then after accumulation are re-weighted by the average of the Gaussian window over the spatial support of the sampling region. The end result is that the extracted features are similar to the exact dense SIFT implementation, but computation is much faster.

Implementation directly based on the VLFeat extractor.

Implementation Notes. The analyser is not thread-safe, however, it is safe to reuse the analyser. In multi-threaded environments, a separate instance must be made for each thread. Internally, this implementation allocates memory for the gradient images, and if possible re-uses these between calls. Re-use requires that the input image is the same size between calls to the analyser.

Author:

Jonathon Hare (jsh2@ecs.soton.ac.uk)

See Also:

"http://www.vlfeat.org/api/dsift.html#dsift-usage"

Field Summary

Fields inherited from class org.openimaj.image.feature.dense.gradient.dsift.DenseSIFT

binHeight, binWidth, data, descriptors, energies, gaussianWindowSize, numBinsX, numBinsY, numOriBins, stepX, stepY, valueThreshold

Constructor Summary

Constructors

Constructor and Description
ApproximateDenseSIFT() Construct with the default configuration: standard SIFT geometry (4x4x8), 5px x 5px spatial bins, 5px step size, gaussian window size of 2 and value threshold of 0.2.
ApproximateDenseSIFT(int step, int binSize) Construct with the given step size (for both x and y) and binSize.
ApproximateDenseSIFT(int stepX, int stepY, int binWidth, int binHeight, int numBinsX, int numBinsY, int numOriBins) Construct with the given configuration.
ApproximateDenseSIFT(int stepX, int stepY, int binWidth, int binHeight, int numBinsX, int numBinsY, int numOriBins, float gaussianWindowSize) Construct with the given configuration.
ApproximateDenseSIFT(int stepX, int stepY, int binWidth, int binHeight, int numBinsX, int numBinsY, int numOriBins, float gaussianWindowSize, float valueThreshold) Construct with the given configuration.

Method Summary

Modifier and Type	Method and Description
ApproximateDenseSIFT	clone()
protected void	extractFeatures() Extract the DSIFT features

Methods inherited from class org.openimaj.image.feature.dense.gradient.dsift.DenseSIFT

analyseImage, getBinHeight, getBinWidth, getByteKeypoints, getByteKeypoints, getDescriptors, getFloatKeypoints, getFloatKeypoints, getNumBinsX, getNumBinsY, getNumOriBins, setBinHeight, setBinWidth

Methods inherited from class org.openimaj.image.feature.dense.gradient.dsift.AbstractDenseSIFT

analyseImage

Methods inherited from class java.lang.Object

equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

ApproximateDenseSIFT

public ApproximateDenseSIFT()

Construct with the default configuration: standard SIFT geometry (4x4x8), 5px x 5px spatial bins, 5px step size, gaussian window size of 2 and value threshold of 0.2.

ApproximateDenseSIFT

public ApproximateDenseSIFT(int step,
                            int binSize)

Construct with the given step size (for both x and y) and binSize. All other values are the defaults.

Parameters:

step - the step size

binSize - the spatial bin size

ApproximateDenseSIFT

public ApproximateDenseSIFT(int stepX,
                            int stepY,
                            int binWidth,
                            int binHeight,
                            int numBinsX,
                            int numBinsY,
                            int numOriBins)

Construct with the given configuration. The gaussian window size is set to 2, and value threshold to 0.2.

Parameters:

stepX - step size in x direction

stepY - step size in y direction

binWidth - width of spatial bins

binHeight - height of spatial bins

numBinsX - number of bins in x direction for each descriptor

numBinsY - number of bins in y direction for each descriptor

numOriBins - number of orientation bins for each descriptor

ApproximateDenseSIFT

public ApproximateDenseSIFT(int stepX,
                            int stepY,
                            int binWidth,
                            int binHeight,
                            int numBinsX,
                            int numBinsY,
                            int numOriBins,
                            float gaussianWindowSize)

Construct with the given configuration. The value threshold is set to 0.2.

Parameters:

stepX - step size in x direction

stepY - step size in y direction

binWidth - width of spatial bins

binHeight - height of spatial bins

numBinsX - number of bins in x direction for each descriptor

numBinsY - number of bins in y direction for each descriptor

numOriBins - number of orientation bins for each descriptor

gaussianWindowSize - the size of the gaussian weighting window

ApproximateDenseSIFT

public ApproximateDenseSIFT(int stepX,
                            int stepY,
                            int binWidth,
                            int binHeight,
                            int numBinsX,
                            int numBinsY,
                            int numOriBins,
                            float gaussianWindowSize,
                            float valueThreshold)

Construct with the given configuration.

Parameters:

stepX - step size in x direction

stepY - step size in y direction

binWidth - width of spatial bins

binHeight - height of spatial bins

numBinsX - number of bins in x direction for each descriptor

numBinsY - number of bins in y direction for each descriptor

numOriBins - number of orientation bins for each descriptor

gaussianWindowSize - the size of the gaussian weighting window

valueThreshold - the threshold for clipping features

Method Detail

extractFeatures

protected void extractFeatures()

Description copied from class: DenseSIFT

Extract the DSIFT features

Overrides:

extractFeatures in class DenseSIFT

clone

public ApproximateDenseSIFT clone()

Overrides:

clone in class DenseSIFT