MeanShift Class 
Namespace: Accord.MachineLearning
[SerializableAttribute] public class MeanShift : ParallelLearningBase, IUnsupervisedLearning<MeanShiftClusterCollection, double[], int>, IClusteringAlgorithm<double[]>, IUnsupervisedLearning<IClusterCollection<double[]>, double[], int>
The MeanShift type exposes the following members.
Name  Description  

MeanShift 
Creates a new MeanShift algorithm.
 
MeanShift(IRadiallySymmetricKernel, Double) 
Creates a new MeanShift algorithm.
 
MeanShift(Int32, IRadiallySymmetricKernel, Double)  Obsolete.
Creates a new MeanShift algorithm.

Name  Description  

Bandwidth 
Gets or sets the bandwidth (radius, or smoothness)
parameter to be used in the meanshift algorithm.
 
Clusters 
Gets the clusters found by Mean Shift.
 
ComputeLabels 
Gets or sets whether cluster labels should be computed
at the end of the learning iteration. Setting to False
might save a few computations in case they are not necessary.
 
ComputeProportions 
Gets or sets whether cluster proportions should be computed
at the end of the learning iteration. Setting to False
might save a few computations in case they are not necessary.
 
Dimension 
Gets the dimension of the samples being
modeled by this clustering algorithm.
 
Distance 
Gets or sets the IMetricT used to
compute distances between points in the clustering.
 
Kernel 
Gets or sets the density kernel to be used in the algorithm.
Default is to use the UniformKernel.
 
Maximum 
Gets or sets the maximum number of neighbors which should be
used to determine the direction of the meanshift during the
computations. Default is zero (unlimited number of neighbors).
 
MaxIterations 
Gets or sets the maximum number of iterations to
be performed by the method. If set to zero, no
iteration limit will be imposed. Default is 0.
 
ParallelOptions 
Gets or sets the parallelization options for this algorithm.
(Inherited from ParallelLearningBase.)  
Token 
Gets or sets a cancellation token that can be used
to cancel the algorithm while it is running.
(Inherited from ParallelLearningBase.)  
Tolerance 
Gets or sets the relative convergence threshold
for stopping the algorithm. Default is 1e3.
 
UseAgglomeration 
Gets or sets whether to use the agglomeration shortcut,
meaning the algorithm will stop early when it detects that
a sample is going to follow the same path as another sample
when running in parallel.
 
UseParallelProcessing  Obsolete.
Gets or sets whether the algorithm can use parallel
processing to speedup computations. Enabling parallel
processing can, however, result in different results
at each run.
 
UseSeeding 
Gets or sets whether to use seeding to initialize the algorithm.
With seeding, new points will be sampled from an uniform grid in
the range of the input points to be used as seeds. Otherwise, the
input points themselves will be used as the initial centroids for
the algorithm.

Name  Description  

Compute(Double)  Obsolete.
Divides the input data into clusters.
 
Compute(Double, Int32)  Obsolete.
Divides the input data into clusters.
 
Equals  Determines whether the specified object is equal to the current object. (Inherited from Object.)  
Finalize  Allows an object to try to free resources and perform other cleanup operations before it is reclaimed by garbage collection. (Inherited from Object.)  
GetHashCode  Serves as the default hash function. (Inherited from Object.)  
GetType  Gets the Type of the current instance. (Inherited from Object.)  
Learn(Double, Double) 
Learns a model that can map the given inputs to the desired outputs.
 
Learn(Double, Int32) 
Learns a model that can map the given inputs to the desired outputs.
 
MemberwiseClone  Creates a shallow copy of the current Object. (Inherited from Object.)  
ToString  Returns a string that represents the current object. (Inherited from Object.) 
Name  Description  

HasMethod 
Checks whether an object implements a method with the given name.
(Defined by ExtensionMethods.)  
IsEqual 
Compares two objects for equality, performing an elementwise
comparison if the elements are vectors or matrices.
(Defined by Matrix.)  
To(Type)  Overloaded.
Converts an object into another type, irrespective of whether
the conversion can be done at compile time or not. This can be
used to convert generic types to numeric types during runtime.
(Defined by ExtensionMethods.)  
ToT  Overloaded.
Converts an object into another type, irrespective of whether
the conversion can be done at compile time or not. This can be
used to convert generic types to numeric types during runtime.
(Defined by ExtensionMethods.) 
Mean shift is a nonparametric featurespace analysis technique originally presented in 1975 by Fukunaga and Hostetler. It is a procedure for locating the maxima of a density function given discrete data sampled from that function. The method iteratively seeks the location of the modes of the distribution using local updates.
As it is, the method would be intractable; however, some clever optimizations such as the use of appropriate data structures and seeding strategies as shown in Lee (2011) and CarreiraPerpinan (2006) can improve its computational speed.
References:
The following example demonstrates how to use the Mean Shift algorithm with a uniform kernel to solve a clustering task:
// Use a fixed seed for reproducibility Accord.Math.Random.Generator.Seed = 0; // Declare some data to be clustered double[][] input = { new double[] { 5, 2, 4 }, new double[] { 5, 5, 6 }, new double[] { 2, 1, 1 }, new double[] { 1, 1, 2 }, new double[] { 1, 2, 2 }, new double[] { 3, 1, 2 }, new double[] { 11, 5, 4 }, new double[] { 15, 5, 6 }, new double[] { 10, 5, 6 }, }; // Create a new MeanShift algorithm for 3 dimensional samples MeanShift meanShift = new MeanShift() { // Use a uniform kernel density Kernel = new UniformKernel(), Bandwidth = 2.0 }; // Learn a data partitioning using the Mean Shift algorithm MeanShiftClusterCollection clustering = meanShift.Learn(input); // Predict group labels for each point int[] labels = clustering.Decide(input); // As a result, the first two observations should belong to the // same cluster (thus having the same label). The same should // happen to the next four observations and to the last three.
The following example demonstrates how to use the Mean Shift algorithm for color clustering. It is the same code which can be found in the color clustering sample application.
// Load a test image (shown in a picture box below) var sampleImages = new TestImages(path: basePath); Bitmap image = sampleImages.GetImage("airplane.png"); // ImageBox.Show("Original", image).Hold(); // Create converters to convert between Bitmap images and double[] arrays var imageToArray = new ImageToArray(min: 1, max: +1); var arrayToImage = new ArrayToImage(image.Width, image.Height, min: 1, max: +1); // Transform the image into an array of pixel values double[][] pixels; imageToArray.Convert(image, out pixels); // Create a MeanShift algorithm using given bandwidth // and a Gaussian density kernel as kernel function. MeanShift meanShift = new MeanShift() { Kernel = new GaussianKernel(3), Bandwidth = 0.06, // We will compute the meanshift algorithm until the means // change less than 0.05 between two iterations of the algorithm Tolerance = 0.05, MaxIterations = 10 }; // Learn the clusters from the data var clusters = meanShift.Learn(pixels); // Use clusters to decide class labels int[] labels = clusters.Decide(pixels); // Replace every pixel with its corresponding centroid double[][] replaced = pixels.Apply((x, i) => clusters.Modes[labels[i]]); // Retrieve the resulting image (shown in a picture box) Bitmap result; arrayToImage.Convert(replaced, out result); // ImageBox.Show("MeanShift clustering", result).Hold();
The original image is shown below:
The resulting image will be: