Tune Registration Settings in Registration Estimator
Adjust settings in Registration Estimator based on your registration technique.
Due to randomness in the registration optimizer, the quality metric, registered image, and geometric transformation can vary slightly between trials despite identical registration settings.
Geometric Transformations Supported by Registration Estimator
All feature-based and intensity-based registration techniques allow you to set the transformation type. For more details about each type of transformation matrix, see Matrix Representation of Geometric Transformations.
Translation transformations preserve the size and orientation of the image. Each pixel in the image is displaced the same amount in the same direction.
Rigid transformations include rotation and translation. Rigid transformations preserve length.
Although reflection is a type of rigid transformation, Registration Estimator does not support reflection.
Similarity transformations include isotropic scaling, rotation, and translation. Similarity transformations preserve shape, but not size. When used with a featured-based registration technique, at least two matched pairs of points are required.
Affine transformations include shear and all supported similarity transformations. Affine transformations preserve parallel lines, but not necessarily angles between lines or distances between points. When used with a featured-based registration technique, at least three matched pairs of points are required.
Projective transformations allow tilting in addition to all supported affine transformations. When used with a featured-based registration technique, at least four matched pairs of points are required.
|All Feature-Based Techniques||X||X||X|
Feature-Based Registration Settings
Feature-based registration allows you to adjust three settings in addition to the geometric transformation type:
Number of detected features. The transformation type determines the minimum number of matched features required to perform a registration. Similarity transformations require two or more matched features. Affine transformations require three or more matched features. Projective transformations require four or more matched features.
Quality of matched features. The quality value is a combination of matched features options.
Rotation. By default, feature-based registration allows the moving image to rotate. However, some imaging scenarios, such as stereoscopy, produce images with identical rotation. If your images have the same rotation, clearing this option can improve the accuracy of the registration.
Intensity-Based Registration Settings
All intensity-based registration techniques allow you to select the geometric transformation type. Additional settings are available depending on the registration technique.
Monomodal and multimodal intensity-based registration provide three common settings:
Normalize. This option scales the pixel values of both images to the same dynamic range.
Apply Gaussian blur. Smoothing the images with a Gaussian blur can help the optimizer find the global maximum or minimum of the solution surface. However, smoothing changes the shape of the surface, and over-smoothing can shift the position of the extrema. Large amounts of blurring are useful when the images are severely misaligned at the start of the registration, to help the optimizer search the correct basin of attraction. Small amounts of blurring are useful when the images start with close alignment.
Align centers. This option provides an initial transformation that aligns the world coordinates of the centers of the two images. The
geometricoption aligns the geometric centers, based on the spatial referencing information of the images. The
center of massoption aligns the centers of mass, calculated from the weighted mean of pixel intensities.
Monomodal registration enables you to adjust the properties of the regular step gradient
descent optimizer. For more information about the properties of this optimizer, see
Multimodal registration enables you to adjust the properties of the one plus one
evolutionary optimizer. For more information about the properties of this optimizer, see
Phase correlation enables you to choose to window the frequency-domain representation of
the images. Windowing increases the stability of registration results. If the common
features you are trying to align in your images are oriented along the edges, clearing this
option can improve registration results. For more information about using phase correlation
to transform an image, see
Nonrigid and Post-Processing Settings
Every registration technique in Registration Estimator allows for nonrigid
transformations to refine the registration fit locally. For more information about
estimating a displacement field for nonrigid transformations, see
The nonrigid settings available in Registration Estimator are:
Number of iterations. This value is the number of iterations on each pyramid level.
Pyramid levels. The value represents the number of Gaussian pyramid reduction levels. The maximum number of pyramid levels depends on the size of each dimension in the images. For example, when the shortest dimension of the fixed and moving images is 256 pixels, at most eight pyramid levels can be used. For more information about pyramid reduction, see
Smoothing. The value represents the standard deviation of Gaussian smoothing and remains the same at each pyramid level. Values are in the range [0.5, 3]. Larger values result in smoother output displacement fields. Smaller values result in more localized deformation in the output displacement field.
Although isotropic scaling and shearing are nonrigid transformations from a mathematical perspective, these transformations act globally on an image. Enable scaling and shearing in the Registration Estimator app by selecting an affine or projective transformation type, not by applying a nonrigid transformation.