TemplateMatching

TemplateMatching is a Julia package designed to offer a native Julia implementation of template matching functionalities similar to those available in OpenCV. This package aims to provide an easy-to-use interface for image processing and computer vision applications, allowing users to leverage the high-performance capabilities of Julia for template matching operations. The package offers performance slightly below that of OpenCV but significantly better than a naive implementation.

Full documentation and description can be found here

To install TemplateMatching, use the Julia package manager. Open your Julia command-line interface and run:

using Pkg
Pkg.add("TemplateMatching")

Masks are not yet supported in the current version of the package. Unlike OpenCV, TemplateMatching.jl supports n-dimensional arrays¹.

Below is a table summarising available methods and their equivalent in opencv.

Full demo can be found here

Import necessary packages

using ImageCore          # Provides core functionalities for image processing
using ImageDraw          # Provides drawing functionalities for images
using TestImages         # Supplies a collection of test images for experimentation

using TemplateMatching

Load the mandrill test image.

img = testimage("mandrill")

Extract a specific portion of the image to use as the template.

template = img[50:80, 150:200]

Convert the image and template to arrays of Float32 type, then perform template matching using Normalized Square Difference as the metric.

img_array = channelview(img) .|> Float32
template_array = channelview(template) .|> Float32

result = match_template(img_array, template_array, NormalizedSquareDiff())
result = dropdims(result, dims = 1)

Display the grayscale version of the result; darker areas indicate closer matches.

result .|> Gray

Identify the location of the best match (the smallest value in the case of Normalized Square Difference), then draw a rectangle around on the original image.

loc = argmin(result)
draw(
    img,
    RectanglePoints(loc[2], loc[1], loc[2] + size(template, 2), loc[1] + size(template, 1)),
    RGB(1, 0, 0)
)

TemplateMatching is provided under the MIT License.