imwarp

Apply geometric transformation to image

Syntax

B = imwarp(A,tform)

B = imwarp(A,D)

[B,RB] = imwarp(A,RA,tform)

[___] = imwarp(___,interp)

[___] = imwarp(___,Name,Value)

Description

B = imwarp(A,tform) transforms the numeric, logical, or categorical image A according to the geometric transformation tform. The function returns the transformed image in B.

B = imwarp(A,D) transforms image A according to the displacement field D.

[B,RB] = imwarp(A,RA,tform) transforms a spatially referenced image specified by the image data A and its associated spatial referencing object RA. The outputs are a spatially referenced image specified by the image data B and its associated spatial referencing object RB.

[___] = imwarp(___,interp) specifies the type of interpolation to use.

[___] = imwarp(___,Name,Value) specifies name-value arguments to control various aspects of the geometric transformation.

Tip

If the input transformation tform does not define a forward transform, then use the OutputView name-value argument to accelerate the transformation.

Examples

Apply Horizontal Shear to Image

Read grayscale image into workspace and display it.

I = imread('cameraman.tif');
imshow(I)

Create a 2-D geometric transformation object.

tform = affine2d([1 0 0; .5 1 0; 0 0 1])

tform = 
  affine2d with properties:

                 T: [3x3 double]
    Dimensionality: 2

Apply the transformation to the image.

J = imwarp(I,tform);
figure
imshow(J)

Apply Rotation Transformation to 3-D MRI Dataset

Read 3-D MRI data into the workspace and visualize it.

s = load('mri');
mriVolume = squeeze(s.D);
sizeIn = size(mriVolume);
hFigOriginal = figure;
hAxOriginal  = axes;
slice(double(mriVolume),sizeIn(2)/2,sizeIn(1)/2,sizeIn(3)/2);
grid on, shading interp, colormap gray

Create a 3-D geometric transformation object. First create a transformation matrix that rotates the image around the y-axis. Then create an affine3d object from the transformation matrix.

theta = pi/8;
t = [cos(theta)  0      -sin(theta)   0
     0             1              0     0
     sin(theta)    0       cos(theta)   0
     0             0              0     1];
tform = affine3d(t)

tform = 
  affine3d with properties:

                 T: [4x4 double]
    Dimensionality: 3

Apply the transformation to the image.

mriVolumeRotated = imwarp(mriVolume,tform);

Visualize three slice planes through the center of the transformed volumes.

sizeOut = size(mriVolumeRotated);
hFigRotated = figure;
hAxRotated  = axes;
slice(double(mriVolumeRotated),sizeOut(2)/2,sizeOut(1)/2,sizeOut(3)/2)
grid on, shading interp, colormap gray

Link the views of both axes together.

linkprop([hAxOriginal,hAxRotated],'View');

Set the view to see the effect of rotation.

set(hAxRotated,'View',[-3.5 20.0])

Warp Image Using Different Output View Styles

Read and display an image. To see the spatial extents of the image, make the axes visible.

A = imread('kobi.png');
iptsetpref('ImshowAxesVisible','on')
imshow(A)

Create a 2-D affine transformation. This example creates a randomized transformation that consists of scale by a factor in the range [1.2, 2.4], rotation by an angle in the range [-45, 45] degrees, and horizontal translation by a distance in the range [100, 200] pixels.

tform = randomAffine2d('Scale',[1.2,2.4],'XTranslation',[100 200],'Rotation',[-45,45]);

Create three different output views for the image and transformation.

centerOutput = affineOutputView(size(A),tform,'BoundsStyle','CenterOutput');
followOutput = affineOutputView(size(A),tform,'BoundsStyle','FollowOutput');
sameAsInput = affineOutputView(size(A),tform,'BoundsStyle','SameAsInput');

Apply the transformation to the input image using each of the different output view styles.

BCenterOutput = imwarp(A,tform,'OutputView',centerOutput);
BFollowOutput = imwarp(A,tform,'OutputView',followOutput);
BSameAsInput = imwarp(A,tform,'OutputView',sameAsInput);

Display the resulting images.

imshow(BCenterOutput)
title('CenterOutput Bounds Style');

imshow(BFollowOutput)
title('FollowOutput Bounds Style');

imshow(BSameAsInput)
title('SameAsInput Bounds Style');

iptsetpref('ImshowAxesVisible','off')

Input Arguments

A — Image to be transformed

numeric array | logical array | categorical array

Image to be transformed, specified as a numeric, logical, or categorical array of any dimension.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64 | logical | categorical

tform — Geometric transformation

rigid2d object | affine2d object | projective2d object | rigid3d object | affine3d object

Geometric transformation to apply, specified as a rigid2d, affine2d, projective2d, rigid3d, or affine3d object.

• If tform is 2-D and A has more than two dimensions, such as for an RGB image, then imwarp applies the same 2-D transformation to all 2-D planes along the higher dimensions.

• If tform is 3-D, then A must be a 3-D image volume.

D — Displacement field

numeric array

Displacement field, specified as numeric array. The displacement field defines the grid size and location of the output image. Displacement values are in units of pixels. imwarp assumes that D is referenced to the default intrinsic coordinate system. To estimate the displacement field, use imregdemons.

• If A is a 2-D grayscale image of size m-by-n, then specify the displacement field as an m-by-n-by-2 array. D(:,:,1) contains displacements along the x-axis. imwarp adds these values to column and row locations in D to produce remapped locations in A. Similarly, D(:,:,2) contains displacements along the y-axis.

• If A is a 2-D RGB or multispectral image of size m-by-n-by-c and you specify D as an m-by-n-by-2 array, then imwarp operates on each 2-D color channel independently. D(:,:,1) contains displacements along the x-axis for all of the color channels. Similarly, D(:,:,2) contains displacements along the y-axis.

• If A is a 3-D grayscale image of size m-by-n-by-p, then specify the displacement field array as an m-by-n-by-p-by-3 array. D(:,:,:,1) contains displacements along the x-axis. imwarp adds these values to column, row, and depth locations in D to produce remapped locations in A. Similarly, D(:,:,:,2) and D(:,:,:,3) contain displacements along the y- and z-axis.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

RA — Spatial referencing information of image to be transformed

imref2d object | imref3d object

Spatial referencing information of the image to be transformed, specified as an imref2d object for a 2-D transformation or an imref3d object for a 3-D transformation.

interp — Type of interpolation used

"nearest" | "linear" | "cubic"

Type of interpolation used, specified as one of these values.

Interpolation Method	Description
"nearest"	Nearest-neighbor interpolation. The output pixel is assigned the value of the pixel that the point falls within. No other pixels are considered. Nearest-neighbor interpolation is the only method supported for categorical images and it is the default method for images of this type.
"linear"	Linear interpolation. This is the default interpolation method for numeric and logical images.
"cubic"	Cubic interpolation

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: B = imwarp(A,tform,FillValues=255) uses a fill value of 255

Before R2021a, use commas to separate each name and value, and enclose Name in quotes.

Example: B = imwarp(A,tform,"FillValues",255) uses a fill value of 255

OutputView — Size and location of output image

imref2d object | imref3d object

Size and location of output image in the world coordinate system, specified as an imref2d or imref3d spatial referencing object. The object has properties that define the size of the output image and the location of the output image in the world coordinate system.

You can create an output view by using the affineOutputView function. To replicate the default output view calculated by imwarp, use the default bounds style ("CenterOutput") of affineOutputView.

You cannot specify OutputView when you specify an input displacement field D.

FillValues — Fill value

numeric scalar | numeric array | string scalar | character vector | missing

Fill values used for output pixels outside the input image, specified as one of the values in the table. imwarp uses fill values for output pixels when the corresponding inverse transformed location in the input image is completely outside the input image boundaries.

The default fill value of numeric and logical images is 0. The default fill value of categorical images is missing, which corresponds to the <undefined> category.

Image Type	Transformation Dimensionality	Format of Fill Values
2-D grayscale or logical image	2-D	Numeric scalar
2-D color image or 2-D multispectral image	2-D	Numeric scalar c-element numeric vector specifying a fill value for each of the c channels. The number of channels, c, is 3 for color images.
Series of p 2-D images	2-D	Numeric scalar c-by-p numeric matrix. The number of channels, c, is 1 for grayscale images and 3 for color images.
N-D image	2-D	Numeric scalar Numeric array whose size matches dimensions 3-to-N of the input image A. For example, if A is 200-by-200-by-10-by-3, then FillValues can be a 10-by-3 array.
3-D grayscale or logical image	3-D	Numeric scalar
Categorical image	2-D or 3-D	Valid category in the image, specified as a string scalar or character vector. missing, which corresponds to the <undefined> category. For more information, see missing.

Example: 255 fills a uint8 image with white pixels

Example: 1 fills a double image with white pixels

Example: [0 1 0] fills a double color image with green pixels

Example: [0 1 0; 0 1 1]', for a series of two double color images, fills the first image with green pixels and the second image with cyan pixels

Example: "vehicle" fills a categorical image with the "vehicle" category

SmoothEdges — Pad image to create smooth edges

false (default) | true

Pad image to create smooth edges, specified as true or false. When set to true, imwarp create a smoother edge in the output image by padding the input image with values specified by FillValues. When set to false, imwarp does not pad the image. Choosing false (not padding) the input image can result in a sharper edge in the output image. This sharper edge can be useful to minimize seam distortions when registering two images side by side.

Output Arguments

B — Transformed image

numeric array | logical array | categorical array

Transformed image, returned as a numeric, logical, or categorical array of the same data type as the input image A.

RB — Spatial referencing information of transformed image

imref2d object | imref3d object

Spatial referencing information of the transformed image, returned as an imref2d or imref3d spatial referencing object.

Algorithms

imwarp determines the value of pixels in the output image by mapping locations in the output image to the corresponding locations in the input image (an inverse mapping). When the center of a pixel in the output image does not map to the center of a pixel in the input image, imwarp interpolates within the input image to calculate the output pixel value.

The figure illustrates a translation transformation of a checkerboard image, in which each square is 10-by-10 pixels. By convention, the axes in input space are labeled u and v and the axes in output space are labeled x and y. Using the inverse transformation, the pixel with (x,y) coordinates (25,35) in the output coordinate space is mapped to the (u,v) coordinates (5,5) in the input coordinate space.

imwarp performs the mapping using world coordinates. For more information, see Image Coordinate Systems.

Extended Capabilities

C/C++ Code Generation

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

Usage notes and limitations:

• imwarp supports the generation of C code (requires MATLAB^® Coder™). Note that if you choose the generic MATLAB Host Computer target platform, imwarp generates code that uses a precompiled, platform-specific shared library. Use of a shared library preserves performance optimizations but limits the target platforms for which code can be generated. For more information, see Types of Code Generation Support in Image Processing Toolbox.

• Input images of data type categorical are not supported.

• The geometric transformation object input, tform, must be a rigid2d, affine2d, or projective2d object.

• The interpolation method and optional parameter names must be constants.

GPU Code Generation

Generate CUDA® code for NVIDIA® GPUs using GPU Coder™.

Usage notes and limitations:

• Input images of data type categorical are not supported.

• The geometric transformation object input, tform, must be a rigid2d, affine2d, or projective2d object.

• The interpolation method and optional parameter names must be constants.

• The spatial referencing information output, RB, is not supported.

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.

GPU Arrays

Accelerate code by running on a graphics processing unit (GPU) using Parallel Computing Toolbox™.

Usage notes and limitations:

• Only "nearest" and "linear" interpolation types are supported.

• Only false is supported for the value of the SmoothEdges argument.

For more information, see Image Processing on a GPU.

Version History

Introduced in R2013a