Mattes mutual information metric configuration
MattesMutualInformation object describes a mutual
information metric configuration that you pass to the function
imregister to solve image registration problems.
You can create a
MattesMutualInformation object using the following
imregconfig— Returns a
MattesMutualInformationobject paired with an appropriate optimizer for registering multimodal images
Enteringon the command line creates a
metric = registration.metric.MattesMutualInformation;
MattesMutualInformationobject with default settings
NumberOfSpatialSamples — Number of spatial samples
500 (default) | positive integer
Number of spatial samples used to compute the mutual information metric,
specified as a positive integer.
defines the number of random pixels
imregister uses to compute
the metric. Your registration results are more reproducible (at the cost of
performance) as you increase this value.
NumberOfHistogramBins — Number of histogram bins
50 (default) | positive integer
Number of histogram bins used to compute the mutual information metric,
specified as a positive integer scalar.
NumberOfHistogramBins defines the number of bins
imregister uses to compute the joint distribution
histogram. The minimum value is
UseAllPixels — Include all pixels in the overlap region
true (default) |
Include all pixels in the overlap region of the images when computing the mutual information metric, specified as a logical scalar.
You can achieve significantly better performance if you set this property
NumberOfSpatialSamples property controls the number
of random pixel locations that
imregister uses to
compute the metric. The results of your registration might not be
UseAllPixels = 0. This is because
imregister selects a random subset of pixels from
the images to compute the metric.
Register Images with Mattes Mutual Information Metric
MattesMutualInformation object and use it to register two MRI images of a knee that were obtained using different protocols.
Read the images into the workspace. The images are multimodal because they have different brightness and contrast.
fixed = dicomread('knee1.dcm'); moving = dicomread('knee2.dcm');
View the misaligned images.
figure imshowpair(fixed, moving,'Scaling','joint');
Create the optimizer configuration object suitable for registering multimodal images.
optimizer = registration.optimizer.OnePlusOneEvolutionary;
Create the metric configuration object suitable for registering multimodal images.
metric = registration.metric.MattesMutualInformation
metric = registration.metric.MattesMutualInformation Properties: NumberOfSpatialSamples: 500 NumberOfHistogramBins: 50 UseAllPixels: 1
Tune the properties of the optimizer so that the problem will converge on a global maxima. Increase the number of iterations the optimizer will use to solve the problem.
optimizer.InitialRadius = 0.009; optimizer.Epsilon = 1.5e-4; optimizer.GrowthFactor = 1.01; optimizer.MaximumIterations = 300;
Perform the registration.
movingRegistered = imregister(moving,fixed,'affine',optimizer,metric);
View the registered images.
figure imshowpair(fixed, movingRegistered,'Scaling','joint');
Larger values of mutual information correspond to better registration results. You can examine the computed values of Mattes mutual information if you enable
DisplayOptimizationwhen you call
imregister, for example:
movingRegistered = imregister(moving,fixed,"rigid",optimizer,metric, ... DisplayOptimization=true);
Mutual information metrics are information theoretic techniques for measuring how related two variables are. These algorithms use the joint probability distribution of a sampling of pixels from two images to measure the certainty that the values of one set of pixels map to similar values in the other image. This information is a quantitative measure of how similar the images are. High mutual information implies a large reduction in the uncertainty (entropy) between the two distributions, signaling that the images are likely better aligned.
The Mattes mutual information algorithm uses a single set of pixel locations for the duration of the optimization, instead of drawing a new set at each iteration. The number of samples used to compute the probability density estimates and the number of bins used to compute the entropy are both user selectable. The marginal and joint probability density function is evaluated at the uniformly spaced bins using the samples. Entropy values are computed by summing over the bins. Zero-order and third-order B-spline kernels are used to compute the probability density functions of the fixed and moving images, respectively .
 Rahunathan, Smriti, D. Stredney, P. Schmalbrock, and B.D. Clymer. Image Registration Using Rigid Registration and Maximization of Mutual Information. Poster presented at: MMVR13. The 13th Annual Medicine Meets Virtual Reality Conference; 2005 January 26–29; Long Beach, CA.
 D. Mattes, D.R. Haynor, H. Vesselle, T. Lewellen, and W. Eubank. "Non-rigid multimodality image registration." (Proceedings paper).Medical Imaging 2001: Image Processing. SPIE Publications, 3 July 2001. pp. 1609–1620.
Run code in the background using MATLAB®
backgroundPool or accelerate code with Parallel Computing Toolbox™
This function fully supports thread-based environments. For more information, see Run MATLAB Functions in Thread-Based Environment.
Version HistoryIntroduced in R2012a
R2021b: Support for thread-based environments
MattesMutualInformation now supports thread-based