Registration Estimator

R2026b

Register 2-D grayscale images

Description

The Registration Estimator app aligns 2-D grayscale images using automatic image registration. Using this app, you can:

• Compare feature-based, intensity-based, and nonrigid registration techniques interactively.

• Obtain the registered image and the geometric transformation.

• Generate a function with the desired registration technique and settings. Call this function to register other images using the same settings.

For more information about the different registration techniques supported by Registration Estimator, see More About.

Open the Registration Estimator App

• MATLAB^® Toolstrip: On the Apps tab, under Image Processing and Computer Vision, click the Registration Estimator app icon.

• MATLAB command prompt: Enter registrationEstimator.

Examples

Load Images and Run Default Registration Trials

Create two misaligned images in the workspace. This example creates the moving image J by rotating the fixed image I clockwise by 30 degrees.

I = imread("cameraman.tif");
J = imrotate(I,-30);

Open Registration Estimator from the command window. Specify the moving image and the fixed image as the two input arguments.

registrationEstimator(J,I)

After you load the images, the app creates three registration trials: Correlation, MSER, and SURF. These trials appear as drafts in the Registrations list. The app also displays an overlay of the images. The default Green-Magenta overlay style shows the fixed image in green and the moving image in magenta. The overlay looks gray in areas where the two images have similar intensity. When you click a feature-based technique in the history list, the image overlay displays the matched features as a set of red and green dots connected by yellow lines.

Run the three default registration trials with the default settings. Click each trial in the history list, then click Register Images.

After the registration finishes, the trial displays a quality score and computation time. The quality score is based loosely on the ssim function and provides an overall estimate of registration quality. A score closer to 1 indicates a higher quality registration.

Inspect the image overlay to confirm which registration technique is the most acceptable. Colors in the image overlay indicate residual misalignment.

Programmatic Use

registrationEstimator opens the Registration Estimator app, which enables you to perform intensity-based, feature-based, and nonrigid image registration.

registrationEstimator(moving,fixed) opens Registration Estimator, loading the grayscale images moving and fixed into the app.

registrationEstimator close closes all open instances of the Registration Estimator app.

More About

Feature-Based Techniques

Feature-based registration techniques automatically detect distinct image features such as blobs, sharp corners, or regions of uniform intensity. Each category has different strengths depending on the content of your images and the expected geometric distortion.

• Use blob detectors when images differ in scale or have significant rotation, with distinct shapes in the scene.

• Use corner detectors when images have scenes with sharp corners, such as aerial views of a city and other constructed scenes. Some corner detectors require the images to have similar scales, such as consecutive video frames or stereoscopic image pairs. Corner detectors provide the fastest results of the feature-based registration techniques.

• Use uniform intensity detectors when images have uniform intensity and features are defined by intensity values. Uniform intensity detectors are more robust to affine transformations than blob detectors and corner detectors.

Technique	Category	Description
SURF	Blob detector	Detect blobs in images with changes in scale or rotation. SURF is faster than SIFT while maintaining good accuracy, making it a good general-purpose registration technique.
SIFT	Blob detector	Detect blobs in images with changes in scale or rotation. SIFT provides the highest accuracy among the blob detectors but is computationally intensive compared to SURF.
ORB	Corner detector	Detect corners in images with changes in scale or rotation. ORB is generally faster and more robust to rotation than BRISK.
FAST	Corner detector	Detect corners in images with little or no scale change. FAST provides a very fast computation, which makes it useful when you need to match a large number of corners in real time. FAST does not support rotation.
BRISK	Corner detector	Detect corners in images with changes in scale or rotation. BRISK generally provides more accurate results for complex scenes than ORB.
Harris	Corner detector	Detect corners in images with little or no scale change. Harris provides a more efficient algorithm than MinEigen, although it can be more sensitive to noise.
MinEigen	Corner detector	Detect corners in images with little or no scale change. MinEigen is the most accurate of the fixed-scale corner detectors, especially when there are strong corners in the images. However, MinEigen is the slowest of the fixed-scale corner detectors.
MSER	Uniform intensity detector	Detect regions of uniform intensity. MSER is more robust to affine transformations than the other feature-based registration algorithms. However, MSER is not suitable when the images differ greatly in intensity.
KAZE	Blob detector	Detect multiscale blobs in images. KAZE is edge-aware, which can improve results near sharp boundaries. KAZE is computationally intensive.

To register the images, the moving image undergoes a single global transformation to provide the best alignment of corresponding features with the fixed image. For more details about each feature detection algorithm and the registration process, see Local Feature Detection and Extraction (Computer Vision Toolbox).

Intensity-Based Techniques

Intensity-based registration techniques correlate image intensity in the spatial or frequency domain. Intensity-based registration can be useful when an image has few distinct shapes or corners, such as for medical images. Registration Estimator offers these registration techniques that use intensity metric optimization.

Technique	Description
Monomodal Intensity	Register images with similar brightness and contrast that are captured on the same type of scanner or sensor. For example, use monomodal intensity to register MRI scans taken of similar subjects using the same imaging sequence.
Multimodal Intensity	Register images with different brightness and contrast. These images can come from two different types of devices, such as two camera models or two types of medical imaging systems (such as CT and MRI). These images can also come from a single device with different settings between captures. For example, use multimodal intensity to register images taken with the same camera using different exposure settings, or to register MRI images acquired during a single session using different imaging sequences.
Correlation	Register images using gradient or phase correlation. Like multimodal intensity, correlation is invariant to image brightness. Correlation is more robust to noise than the other intensity-based registration techniques.

To register the images, the moving image undergoes a single global transformation to maximize the correlation of its intensity with the intensity of the fixed image. For more information about the registration process, see Intensity-Based Automatic Image Registration.

Nonrigid Registration Techniques

Nonrigid transformations generate a displacement field, in which each pixel location in the fixed image is mapped to a corresponding location in the moving image. The moving image is then warped according to the displacement field and resampled using linear interpolation. For more information about estimating a displacement field for nonrigid transformations, see imregdemons.

Technique	Description
Nonrigid	Register images with local distortions.

Tips

• You can register images and generate functions for all feature-based techniques without a Computer Vision Toolbox™ license. However, to run an autogenerated function that uses a feature-based registration technique, you must have Computer Vision Toolbox. For more information, see Export Results from Registration Estimator App.

Version History

Introduced in R2017a

R2026b: Support for SIFT registration technique

The app now supports the SIFT feature-based registration technique. You can use SIFT to register images with rotation and scale differences. SIFT provides the highest accuracy among the feature-based blob detectors, but it is computationally intensive compared to SURF.

R2021a: Support for KAZE and ORB registration techniques

The app now supports the KAZE and ORB feature-based registration techniques. The KAZE technique is a multiscale 2-D feature detection and description algorithm in nonlinear scale spaces. The ORB technique detects corners in images with changes in scale, rotations, or both.