Image processing: Average grayscale images

Question

I would like to average two grayscale images (both png files of the same size) using R.

Consider the images below as an example:

img1.png

img2.png

img3.png

The average of img1.png and img2.png should result in something like img3.png. I've looked into the magick and imager packages without success.

Answer 1

Surely there are multiple solutions to this problem. Here's one that shows clearly what's happening and allows for more flexibility.

Let's load your two example pictures.

library(png)
img1 <- readPNG("owzUE.png")
img2 <- readPNG("HMka0.png")

It happens that they are not of identical size:

dim(img1)
# [1] 145 190   4
dim(img2)
# [1] 144 191   4

So, let's make the size the same with

img1 <- img1[1:144, 1:190, 1:4]
img2 <- img2[1:144, 1:190, 1:4]

Each image consists of four matrices, the first three corresponding to the RGB channels and the last one to alpha. Let's define the resulting image by

img3 <- img1 * 0

which nicely gives an array of the same size as img1 and img2.

A closely related concept is alpha blending but the formulas are not exactly what you need; there the relationship between images doesn't seem to be symmetric. What we may do is to consider a weighted average for each RGB channel, where weights correspond to the corresponding alphas:

for(d in 1:3)
  img3[, , d] <- (img1[, , d] * img1[, , 4]  + img2[, , d] * img2[, , 4] ) / (img1[, , 4] + img2[, , 4])

Since in some cases both alphas are 0, we fix those cases with

img3[is.nan(img3)] <- 0

Lastly, we set alpha value to a larger one of the two images with

img3[, , 4] <- pmax(img1[, , 4], img2[, , 4])

and save the result

writePNG(img3, "img3.png")

So, this method also applies to color images and allows one to experiment with the formula by varying weights, etc.