Packages

p

scalismo

registration

package registration

Ordering
  1. Alphabetic
Visibility
  1. Public
  2. Protected

Type Members

  1. case class GaussianProcessTransformation[D](gp: LowRankGaussianProcess[D, EuclideanVector[D]], alpha: ParameterVector)(implicit vectorizer: Vectorizer[EuclideanVector[D]]) extends ParametricTransformation[D] with Product with Serializable
  2. case class GaussianProcessTransformationSpace[D](gp: LowRankGaussianProcess[D, EuclideanVector[D]])(implicit vectorizer: VectorVectorizer[D]) extends TransformationSpace[D] with Product with Serializable
  3. trait ImageMetric[D, A] extends RegistrationMetric[D]

    A registration metric for image to image registration.

  4. case class L2Regularizer[D](transformationSpace: TransformationSpace[D]) extends Regularizer[D] with Product with Serializable

    A regularizer which simply penalizes the squared norm of the parameters.

  5. case class MeanHuberLossMetric[D, A](fixedImage: Field[D, A], movingImage: DifferentiableField[D, A], transformationSpace: TransformationSpace[D], sampler: Sampler[D], delta: Double = 1.345)(implicit evidence$1: NDSpace[D], evidence$2: Scalar[A]) extends MeanPointwiseLossMetric[D, A] with Product with Serializable

    Image to image metric which applies the Huber Loss function to the pointwise pixel difference.

    Image to image metric which applies the Huber Loss function to the pointwise pixel difference. The parameter delta defines the threshold. The Huber loss increases quadratically for values below this threshold and linearly for values above this threshold.

    See also

    SumOfPointwiseLossMetric.

  6. abstract class MeanPointwiseLossMetric[D, A] extends ImageMetric[D, A]

    Image to image metric which applies a loss function to the pointwise pixel difference.

    Image to image metric which applies a loss function to the pointwise pixel difference. The total value of the metric is the mean of this pointwise loss. The points are determined by the sampler.

  7. case class MeanSquaresMetric[D, A](fixedImage: Field[D, A], movingImage: DifferentiableField[D, A], transformationSpace: TransformationSpace[D], sampler: Sampler[D])(implicit evidence$1: NDSpace[D], evidence$2: Scalar[A]) extends MeanPointwiseLossMetric[D, A] with Product with Serializable

    The mean squares image to image metric.

    The mean squares image to image metric. It is implemented as the squared loss function in terms of the pointwise pixel differences.

  8. case class MutualInformationMetric[D, A](fixedImage: Field[D, A], fixedImageDomain: DiscreteImageDomain[D], movingImage: DifferentiableField[D, A], transformationSpace: TransformationSpace[D], sampler: Sampler[D], numberOfBins: Int = 30)(implicit evidence$1: NDSpace[D], evidence$2: Scalar[A], rng: Random) extends ImageMetric[D, A] with Product with Serializable

    Implementation of the Mutual Information Metric, described in the following paper:

    Implementation of the Mutual Information Metric, described in the following paper:

    Mattes, David, et al. "PET-CT image registration in the chest using free-form deformations." IEEE transactions on medical imaging 22.1 (2003): 120-128.

    fixedImage

    The fixed image

    fixedImageDomain

    The domain of the fixed image. All grid points of the domain are used to compute image characteristics such as e.g. the minimum/maximum value, etc.

    movingImage

    The moving image

    transformationSpace

    The transformation space that is used

    sampler

    The sampler, which samples the points on which the mutual information is computed. For this metric the recommended choice is a random sampler (which combined with a gradient descent algorithm leads to a stochastic gradient descent.

    numberOfBins

    The number of bins used for the intensity histograms (which approximates the joint distribution)

  9. case class Registration[D](metric: RegistrationMetric[D], regularizer: Regularizer[D], regularizationWeight: Double, optimizer: Optimizer) extends Product with Serializable

    Implementation of a gradient-based registration algorithm, whose cost function is defined by the sum of a distance metric and a regularization term (weighted by a regularizationWeight).

    Implementation of a gradient-based registration algorithm, whose cost function is defined by the sum of a distance metric and a regularization term (weighted by a regularizationWeight).

    metric

    The distance metric used to compare the objects that should be registered.

    regularizer

    The regularizer that is used

    regularizationWeight

    A weight used to weight the influence of the regularizer (0 means the regularization term is not considered)

    optimizer

    The optimizer used to perform the minimization of the cost function

  10. trait RegistrationMetric[D] extends AnyRef

    The basic interface for defining a metric for the scalismo registration framework.

    The basic interface for defining a metric for the scalismo registration framework. It is independent of the object representation that is used. The main assumption is that the objects are subject to a parametric transformation, and we can obtain for each fixed value of the parameters compute the value of the metric and a derivative.

  11. trait Regularizer[D] extends AnyRef

    This trait defines the interface for regularizers in the scalismo registration framework.

    This trait defines the interface for regularizers in the scalismo registration framework. A regularizer is defined as a differentiable functions of the registration parameters.

Ungrouped