In battmanux/cv2r: Use OpenCV in RStudio

cv2r is an R wrapper to Pyhon OpenCV using reticulate. There is a few additions to simplify the integration in RStudio.

The very first time you may need to install OpenCV lib in your python environment.

library(reticulate)
library(cv2r)
install_opencv()

Than it is pretty simle to read and show an image.

img_url <- "https://upload.wikimedia.org/wikipedia/fr/4/4e/RStudio_Logo.png"

my_image <- imread(img_url)

# Very simple to plot images
imshow(mat=my_image)

There is also a few modifications to simplify image pixel selection. You can use R matrix index on an numpy.ndarray variable.

# Lets show the alpha mask
imshow(mat=my_image[,,4])

# To use advances matrix subseting you shall convert the matrix to R 
my_r_image <- reticulate::py_to_r(my_image)
imshow(mat=my_r_image[50:300,200:900,c(4,2,2)])

If you are using RStudio 1.2 you can call plot from a python chunk

print(r.my_image.shape)

You can naturraly do your cv2 code in python

import cv2

blured_img = cv2.blur(r.my_image, (100,100) )

# We call the cv2r::imshow version to get the result in the markdown document
r.imshow(mat=blured_img)

You can change the colorspace

# change color space
cvtColor(my_image) <- "HSV"

# change Hue
my_image[,,1] <- my_image[,,1]*2+50

imshow(mat=my_image)

Convert to data.table

library(data.table)

my_image <- imread(img_url)

my_table <- as.data.table(my_image)

summary(my_table)

So that you can do data science on pixels

hist(my_table[, R])

Then change pixels and revert to image

imshow(mat=my_table)

If you need to put in a 3D space

pairs(my_table[sample.int(size = 100, n =  nrow(my_table)),.(R,G,B, A)])

Than you update values in 3D spaces

my_table[R < 100 & G > 50, R:=200]
imshow(mat=my_table)

You can also remove points in any colorspace and plot the result with transparency

# remove alpha before converting to HSV
my_hsv_image <- my_image[,,1:3] 
cvtColor(my_hsv_image) <- "HSV"
my_hsv_table <- as.data.table(my_hsv_image)
imshow(mat=my_hsv_table[V > 200 & V < 240,])