Image Convolution in R using Magick

[This article was first published on rOpenSci - open tools for open science, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Release 1.4 of the magick package introduces a new feature called image convolution that was requested by Thomas L. Pedersen. In this post we explain what this is all about.

Kernel Matrix

The new image_convolve() function applies a kernel over the image. Kernel convolution means that each pixel value is recalculated using the weighted neighborhood sum defined in the kernel matrix. For example lets look at this simple kernel:


kern <- matrix(0, ncol = 3, nrow = 3)
kern[1, 2] <- 0.25
kern[2, c(1, 3)] <- 0.25
kern[3, 2] <- 0.25
##      [,1] [,2] [,3]
## [1,] 0.00 0.25 0.00
## [2,] 0.25 0.00 0.25
## [3,] 0.00 0.25 0.00

This kernel changes each pixel to the mean of its horizontal and vertical neighboring pixels, which results in a slight blurring effect in the right-hand image below:

img <- image_read('logo:')
img_blurred <- image_convolve(img, kern)
image_append(c(img, img_blurred))


Standard Kernels

Many operations in magick such as blurring, sharpening, and edge detection are actually special cases of image convolution. The benefit of explicitly using image_convolve() is more control. For example, we can blur an image and then blend it together with the original image in one step by mixing a blurring kernel with the unit kernel:

img %>% image_convolve('Gaussian:0x5', scaling = '60,40%')


The above requires a bit of explanation. ImageMagick defines several common standard kernels such as the gaussian kernel. Most of the standard kernels take one or more parameters, e.g. the example above used a gaussian kernel with 0 radius and 5 sigma.

In addition, scaling argument defines the magnitude of the kernel, and possibly how much of the original picture should be mixed in. Here we mix 60% of the blurring with 40% of the original picture in order to get a diffused lightning effect.

Edge Detection

Another area where kernels are of use is in edge detection. A simple example of a direction-aware edge detection kernel is the Sobel kernel. As can be seen below, vertical edges are detected while horizontals are not.

img %>% image_convolve('Sobel') %>% image_negate()


Something less apparent is that the result of the edge detection is truncated. Edge detection kernels can result in negative color values which get truncated to zero. To combat this it is possible to add a bias to the result. Often you’ll end up with scaling the kernel to 50% and adding 50% bias to move the midpoint of the result to 50% grey:

img %>% image_convolve('Sobel', scaling = '50%', bias = '50%')



ImageMagick has many more edge detection kernels, some of which are insensitive to the direction of the edge. To emulate a classic high-pass filter from photoshop use difference of gaussians kernel:

img %>% image_convolve('DoG:0,0,2') %>% image_negate()


As with the blurring, the original image can be blended in with the transformed one, effectively sharpening the image along edges.

img %>% image_convolve('DoG:0,0,2', scaling = '100, 100%')


The ImageMagick documentation has more examples of convolve with various avaiable kernels.

To leave a comment for the author, please follow the link and comment on their blog: rOpenSci - open tools for open science. offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)