imregcorr

Estimate geometric transformation that aligns two 2-D images using cross correlation

Syntax

tform = imregcorr(moving,fixed)

tform = imregcorr(moving,Rmoving,fixed,Rfixed)

tform = imregcorr(___,tformType)

tform = imregcorr(___,Name=Value)

[tform,peakcorr] = imregcorr(___)

Description

tform = imregcorr(moving,fixed) estimates the geometric transformation that aligns an image, moving, with a reference image, fixed, using gradient correlation.

tform = imregcorr(moving,Rmoving,fixed,Rfixed) estimates the geometric transformation that aligns an image, moving, with a reference image, fixed. Rmoving and Rfixed are spatial referencing objects that contain spatial information about the moving and fixed images, respectively. The geometric transformation, tform, defines the point mapping in the world coordinate system.

tform = imregcorr(___,tformType) also specifies the type of transformation, tformType.

tform = imregcorr(___,Name=Value) enables you to specify the cross correlation method as phase correlation, and to enable or disable windowing for phase correlation, using name-value arguments.

[tform,peakcorr] = imregcorr(___) also returns the peak correlation value, peakcorr.

Examples

Register Images Using Phase Correlation

Read a reference image into the workspace.

fixed  = imread("cameraman.tif");

Create a synthetic moving image by scaling and rotating the fixed image.

scaleFactor = 2.3;
theta = 20;
translation = [0 0];
tform = simtform2d(scaleFactor,theta,translation);
moving = imwarp(fixed,tform);

Add synthetic noise to the moving image.

moving = moving + uint8(10*rand(size(moving)));

Display the fixed and the moving image alongside each other.

imshowpair(fixed,moving,"montage")

Estimate the transformation needed to align the images using imregcorr.

tformEstimate = imregcorr(moving,fixed);

Apply the estimated geometric transform to the moving image. Specify the "OutputView" name-value argument to obtain a registered image the same size and with the same world limits as the reference image.

Rfixed = imref2d(size(fixed));
movingReg = imwarp(moving,tformEstimate,"OutputView",Rfixed);

View the original image and the registered image side-by-side to check the registration. Then view the registered image overlaid on the original using the "falsecolor" option to highlight any areas where the images differ.

imshowpair(fixed,movingReg,"montage")

imshowpair(fixed,movingReg,"falsecolor");

Input Arguments

moving — Moving image

grayscale image | binary image | RGB image

Moving image, or the image to be registered, specified as a grayscale, binary, or RGB image. imregcorr converts RGB images to grayscale using rgb2gray before processing.

Data Types: single | double | int8 | int16 | int32 | uint8 | uint16 | uint32 | logical

fixed — Reference image

grayscale image | binary image | RGB image

Reference image in the target orientation, specified as a grayscale, binary, or RGB image. imregcorr converts RGB images to grayscale using rgb2gray before processing.

Data Types: single | double | int8 | int16 | int32 | uint8 | uint16 | uint32 | logical

Rmoving — Spatial referencing information of moving image

imref2d object

Spatial referencing information of the moving image, specified as an imref2d object.

Rfixed — Spatial referencing information of fixed image

imref2d object

Spatial referencing information of the fixed image, specified as an imref2d object.

tformType — Type of transformation to estimate

"similarity" (default) | "rigid" | "translation"

Type of transformation to estimate, specified as one of these values.

Value	Description
"translation"	Translation transformation When using the "translation" option with spatial referencing objects Rmoving and Rfixed, the input images must have the same pixel extents in world coordinates.
"rigid"	Rigid transformation: translation and rotation When using the "rigid" option with spatial referencing objects Rmoving and Rfixed, the input images must have the same pixel extents in world coordinates.
"similarity"	Similarity transformation: translation, rotation, and isotropic scaling When using the "similarity" option with the phase correlation method, imregcorr does not detect scale differences less than 1/4 or greater than 4.

Data Types: char | string

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Example: tform = imregcorr(moving,fixed,Method="phasecorr") uses phase correlation.

Method — Correlation method

"gradcorr" (default) | "phasecorr"

Since R2024b

Correlation method, specified as "gradcorr" for gradient correlation or "phasecorr" for phase correlation. If either input image is a vector, then use phase correlation.

Window — Use windowing to suppress spectral leakage effects

true or 1 (default) | false or 0

Use windowing to suppress spectral leakage effects in the frequency domain, specified as a numeric or logical 1 (true) or 0 (false). When Window is true and Method is "phasecorr", the imregcorr function performs windowing using a Blackman filter. The function ignores the Window argument when Method is "gradcorr".

To increase the stability of registration results for phase correlation, specify Window as true. However, if the common features in your images are oriented along the edges, then setting Window to false can sometimes provide superior registration results.

Output Arguments

tform — Geometric transformation

transltform2d object | rigidtform2d object | simtform2d object

Geometric transformation, returned as a geometric transformation object according to the type of transformation, tformType.

tformType	Geometric Transformation Object
"translation"	transltform2d
"rigid"	rigidtform2d
"similarity"	simtform2d

peakcorr — Peak correlation

number in the range [0, 1]

Peak correlation value between the two images, returned as a number in the range [0, 1]. A larger peak correlation value indicates more confidence in the estimated geometric transformation.

Tips

• If your image is of type double, you can achieve performance improvements by casting the image to single with im2single before registration. Input images of type double cause the algorithm to compute FFTs in double.

• Because of differences in the gradient correlation and phase correlation algorithms, the peak correlation value peakcorr should not be used as a metric to compare the performance of the algorithms.

References

[1] Reddy, B. S. and Chatterji, B. N. "An FFT-Based Technique for Translation, Rotation, and Scale-Invariant Image Registration." IEEE Transactions on Image Processing, Vol. 5, No. 8, August 1996.

[2] Tzimiropoulos, Georgios, Vasileios Argyriou, Stefanos Zafeiriou, and Tania Stathaki. “Robust FFT-Based Scale-Invariant Image Registration with Image Gradients.” IEEE Transactions on Pattern Analysis and Machine Intelligence 32, no. 10 (October 2010): 1899–1906. https://doi.org/10.1109/TPAMI.2010.107.

Extended Capabilities

C/C++ Code Generation

Generate C and C++ code using MATLAB® Coder™.

Usage notes and limitations:

• imregcorr supports the generation of C and C++ code (requires MATLAB^® Coder™). For more information, see Code Generation for Image Processing.

• The optional tformType, Method, and Window arguments must be compile-time constants.

Thread-Based Environment

Run code in the background using MATLAB® backgroundPool or accelerate code with Parallel Computing Toolbox™ ThreadPool.

This function fully supports thread-based environments. For more information, see Run MATLAB Functions in Thread-Based Environment.

Version History

Introduced in R2014a

R2024b: Performs gradient correlation by default

Starting in R2024b, the imregcorr function uses a new default method called gradient correlation. In many situations, gradient correlation is more accurate and faster than phase correlation. The gradient correlation algorithm is based on an algorithm by Tzimiropoulos [2], with some modifications.

Prior to R2024b, imregcorr supported only the phase correlation method. To use the phase correlation method, specify the new Method name-value argument as "phasecorr".

R2024b: Improved phase correlation algorithm

The imregcorr function implements an improved phase correlation algorithm that returns more accurate and consistent results. The improvements can cause different behavior from prior releases, such as:

• A larger value for the peak correlation.

• A modified window when performing windowing in the frequency domain. The new behavior is most noticeable when the input images have a small number of rows or columns.

• More consistent windowing when estimating rigid and similarity transformations.

Now, no warning is issued when the peak correlation value is less than 0.03.

R2022b: Returns new geometric transformation objects

Starting in R2022b, most Image Processing Toolbox™ functions create and perform geometric transformations using the premultiply convention. Accordingly, the imregcorr function now returns the output argument tform as an transltform2d, rigidtform2d, or simtform2d object, which support the premultiply convention. Before, the function returned tform as an affine2d object, which supports the postmultiply convention. For more information, see Migrate Geometric Transformations to Premultiply Convention.

R2021b: Support for thread-based environments

imregcorr now supports thread-based environments.