Generate Code to Resize Images to Fixed Output Size

Scenario 1: Generate Code to Resize Constant Input Image

In this scenario, you want to generate code that resizes a single image with a fixed filename. The input image is a constant input to the codegen function because the value of the argument is always the same. Specify the filename of the image.

imageFileName = "foggyroad.jpg";

Define Entry-Point Function

The MATLAB function that you want to convert is called the entry-point function. In this scenario, the entry-point function is a custom function called imresizeConstant. The imresizeConstant function reads a single image from a file and then resizes the image using the imresize function. The function supports both rescaling by an isotropic scale factor and resizing both image dimensions to a target size.

type imresizeConstant.m

function out = imresizeConstant(fileName,scaleOrSize)
% Copyright 2022 The MathWorks, Inc.

% Reads the image using imread
I = imread(fileName);

% Resize the image using Scale or Size
out = imresize(I,scaleOrSize);
end

Option 1: Resize Using Constant Scale Factor

Specify the scale factor.

scaleFactor = 0.5;

Generate the MEX file using the codegen function. You must define the data type of each input argument of the entry-point function at compile time because the C and C++ coding languages use static typing. For this scenario, you can further specify both input arguments as constant values using the coder.Constant class. For more information, see Specify Types of Entry-Point Function Inputs (MATLAB Coder).

codegen -config:mex imresizeConstant.m -args {coder.Constant(imageFileName),coder.Constant(scaleFactor)} -report

Code generation successful: To view the report, open('codegen/mex/imresizeConstant/html/report.mldatx')

To open the code generation report, click View report. In the Function pane, the orange font color of both input arguments indicates that they are constant arguments to the entry-point function. If you pause on the out variable, the Variable Info tooltip indicates that out has a fixed size, as expected. The size is equal to one half of the input size. Because the input image does not change, the output image size is consistent each time you run the MEX file.

Option 2: Resize Using Constant Target Size

Specify a constant target size.

targetSize = [256 256];

Generate the MEX file using the codegen function. Define the input argument types, and use the coder.Constant class to specify both inputs as constant values. The compiled function resizes the input image to a target size because targetSize is a vector.

codegen -config:mex imresizeConstant.m -args {coder.Constant(imageFileName),coder.Constant(targetSize)} -report

Code generation successful: To view the report, open('codegen/mex/imresizeConstant/html/report.mldatx')

Open the code generation report. In the Function pane, the orange font color indicates that both inputs are constant. If you pause on the out variable, the Variable Info tooltip indicates that out has a fixed size, equal to the target size specified by targetSize.

Scenario 2: Generate Code to Resize Variable Input Image

In this scenario, you want to generate code that accepts any input image filename for resizing. The input image is a variable input to the codegen function because the value of the argument can change.

To ensure that different input images always have the same output size, you must resize using a constant target size. Rescaling by a scale factor does not guarantee that the output images will have a fixed size, because input images with different sizes will yield different output sizes.

Define Entry-Point Function

For this scenario, define a new entry-point function imresizeVariable.m. This function ensures that the resized image has three channels. If the input image is a grayscale image, the function concatenates three copies of the image along the third dimension. If the input image is a multispectral image or volume, the function preserves only the first three channels.

type imresizeVariable.m

function out = imresizeVariable(fileName,targetSize)
% Copyright 2022 The MathWorks, Inc.

% Read the image using imread
I = imread(fileName);

% Ensure the image has three channels by concatenating the channels of a
% grayscale image and taking the first three channels of a multispectral
% image
if (size(I,3)<3)
    I = cat(3,I,I,I);
end
I = I(:,:,1:3);

% Resize the image to target size
out = imresize(I,targetSize);
end

Resize Using Constant Target Size

Specify a constant target size to use as input to the entry-point function.

targetSize = [256 256];

Generate the MEX file using the codegen function. Specify a variable filename of data type string, and a coder.Constant class for the target output size.

codegen -config:mex imresizeVariable.m -args {imageFileName,coder.Constant(targetSize)} -report

Code generation successful: To view the report, open('codegen/mex/imresizeVariable/html/report.mldatx')

Open the code generation report. In the Function pane, the orange font color indicates that only targetSize is constant, while fileName can vary. If you pause on the out variable, the Variable Info tooltip indicates that all three dimensions of the output image are fixed, as desired.

Note that if you accidentally specify a scale factor instead of a target size, the Variable Info tooltip displays a question mark for one or more of the dimensions. The question mark indicates that the size of the output image is not fixed.

Scenario 3: Generate Code to Resize Multiple Input Images

In this scenario, you want to generate code that accepts multiple input image filenames for resizing. The input image is a variable input to the codegen function because the value of the argument can change.

To ensure that different input images always have the same output size, you must resize using a constant target size. Specify the filenames of the image.

imageFileName1 = "foggyroad.jpg";
imageFileName2 = "foosball.jpg";
imFileNames = {imageFileName1,imageFileName2};

Define Entry-Point Function

Define an entry-point function named imresizeVariableBatch.m that reads a batch of input images specified as a cell array, and resizes each image to a target size. The function returns the resized images as a 4-D array, where the fourth dimension corresponds to the number of input images.

type imresizeVariableBatch.m

function out = imresizeVariableBatch(imageFileNameCellArray,targetSize)
% Copyright 2022 The MathWorks, Inc.

numImages = coder.const(numel(imageFileNameCellArray));

% Allocate memory for output image
out = coder.nullcopy(zeros([targetSize,3,numImages],"uint8"));

% For each image in the input cell array, read and resize the image.
for i = coder.unroll(1:numImages)
    I = imread(imageFileNameCellArray{i});

    if (size(I,3)<3)
        I = cat(3,I,I,I);
    end
    I = I(:,:,1:3);

    out(:,:,:,i) = imresize(I,targetSize);
end
end

Resize Using Constant Target Size

Specify a constant target size to use as input to the entry-point function.

targetSize = [256 256];

Generate the MEX file using the codegen function. Specify the first input argument as a cell array containing the filenames. The codegen function configures the MEX file to accept a cell array of two strings, which can vary. Specify a coder.Constant class for the target output size.

codegen -config:mex imresizeVariableBatch.m -args {imFileNames,coder.Constant(targetSize)} -report

Code generation successful: To view the report, open('codegen/mex/imresizeVariableBatch/html/report.mldatx')

Open the code generation report. In the Function pane, the orange font color indicates that only targetSize is constant, while imageFileNameCellArray can vary. If you pause on the out variable, the Variable Info tooltip indicates that all dimensions of the 4-D output image array are fixed.