Main Content

Choose an Image Registration Technique

There are several factors to consider when selecting an approach to register images:

  • Some techniques are automated. Other techniques are interactive to support exploration, or semi-automated to enable precise control.

  • Some techniques require the images to be related by a rigid geometric transformation. Other techniques are more flexible and support projective transformations, local deformations, or other types of distortion. Select a technique capable of handling the expected transformation model.

  • Some techniques support only 2-D images, whereas others also support 3-D image volumes.

GoalApproach to Register ImageSample Display
Interactively explore, compare, and tune a variety of registration techniques for 2-D images.

Use the Registration Estimator app. The app supports eight feature-based techniques, three intensity-based techniques, and one nonrigid (deformable) technique.

The app provides a quantitative measure of quality, and it returns the registered image and the transformation matrix. The app can also generate code with your selected registration technique and settings.

See Register Images Using Registration Estimator App.

Registration Estimator app showing an overlay of images with corresponding control points.

Automatically register 2-D images based on image features.

Use automated feature matching. Automated feature matching is a powerful registration technique that can find image correspondences regardless of occlusion, changes in viewing conditions, or the presence of clutter. Automated feature matching is used in a wide range of applications such as video stabilization and panoramic image stitching.

You can pick one or more feature types and descriptors based on the content of the image. For blob features, common feature types include the scale invariant feature transform (SIFT) and Speeded-Up Robust Features (SURF). For fixed-scale corner features, a common feature type is Oriented FAST and rotated BRIEF (ORB).

For a list of all available feature detectors and examples that demonstrate how to perform automated feature detection, see Local Feature Detection and Extraction (Computer Vision Toolbox).

Matched features using a BRISK feature detector.

Automatically register 2-D or 3-D grayscale images based on relative intensity patterns.

Use intensity-based image registration techniques. These techniques optimize the similarity of pixel intensities in the two images based on a metric.

  • imregcorr quickly optimizes the similarity using gradient or phase cross-correlation.

  • imregister performs a fine-tuned optimization by using a mean square or Mattes mutual information optimizer. This optimization can be more accurate than cross-correlation, but might be slower or require more tuning of optimization parameters.

  • imregicp (Medical Imaging Toolbox) aligns surfaces extracted from 3-D grayscale volumes.

See Intensity-Based Automatic Image Registration.

Intensity-based registration using cross-correlation aligns images that have shear distortion.

Automatically perform a deformable (nonrigid) registration.

Use the Registration Estimator app to interactively tune the registration settings for deformable registration.

Alternatively, you can estimate a localized displacement field by using the imregdemons function. This function returns the displacement field and the registered moving image.

Deformable registration aligns two hands whose fingers are spread to different extents.

Manually register 2-D images in situations where other registration techniques are unsupported or return poor results.

Use control point registration, which enables you to select common features in each image manually. Control point registration is useful when:

  • Automated feature matching algorithms cannot identify clear mappings between features.

  • The moving image has a non-affine geometric transformation, such as a projective, polynomial, piecewise linear, or local weighted mean transformation.

See Control Point Registration and Register Images with Projection Distortion Using Control Points.

Nine pairs of control points selected in the Control Point Selection tool.

Align medical images and volumes in a common patient coordinate system.

Medical Imaging Toolbox™ provides additional techniques and examples for registering 2-D and 3-D medical image data in the patient coordinate system.

  • Interactively register 3-D medical image volumes by using the Medical Registration Estimator (Medical Imaging Toolbox) app. The app automatically applies metadata from medical file formats to easily register volumes in patient coordinates.

  • Use automated intensity-based optimization techniques tailored to 2-D or 3-D medical data. For example, the imreggroupwise (Medical Imaging Toolbox) function reduces sliding motion between slices in a series of 2-D images.

  • Perform 3-D control point registration by using the fitgeotform3d (Medical Imaging Toolbox) function.

See Medical Image Registration (Medical Imaging Toolbox).

Medical Registration Estimator app window

See Also

|

Related Topics