extract RGB channels from a jpeg image in R

Question

In order to classify a jpeg image in R, I would like to get the RGB values of each pixel.

My question: Is there a way to extract RGB channels from a jpeg image in R ?

Answer 1

You have several package to read in JPEG. Here I use package jpeg:

library(jpeg)
img <- readJPEG("Rlogo.jpg")

dim(img)
[1]  76 100   3

As you can see, there is 3 layers: they correspond to your R, G and B values. In each layer, each cell is a pixel.

img[35:39,50:54,]
, , 1

          [,1]      [,2]      [,3]      [,4]      [,5]
[1,] 0.5098039 0.5921569 0.4549020 0.3372549 0.1921569
[2,] 0.5098039 0.6000000 0.4549020 0.3372549 0.1921569
[3,] 0.5137255 0.6000000 0.4549020 0.3450980 0.1921569
[4,] 0.5215686 0.6039216 0.4627451 0.3450980 0.1921569
[5,] 0.5215686 0.6039216 0.4627451 0.3450980 0.1882353

, , 2

          [,1]      [,2]      [,3]      [,4]      [,5]
[1,] 0.5882353 0.6666667 0.5098039 0.3803922 0.2156863
[2,] 0.5882353 0.6627451 0.5098039 0.3803922 0.2156863
[3,] 0.5843137 0.6627451 0.5098039 0.3764706 0.2156863
[4,] 0.5843137 0.6627451 0.5058824 0.3764706 0.2117647
[5,] 0.5843137 0.6627451 0.5058824 0.3764706 0.2156863

, , 3

          [,1]      [,2]      [,3]      [,4]      [,5]
[1,] 0.7254902 0.7921569 0.6156863 0.4588235 0.2705882
[2,] 0.7254902 0.7921569 0.6156863 0.4588235 0.2784314
[3,] 0.7254902 0.7921569 0.6156863 0.4588235 0.2784314
[4,] 0.7176471 0.7921569 0.6156863 0.4666667 0.2862745
[5,] 0.7176471 0.7921569 0.6156863 0.4666667 0.2862745

Answer 2

I recommend the biOpspackage for image manipulation.

Here is an example:

library(biOps)
x <- readJpeg(system.file("samples", "violet.jpg", package="biOps"))
plot(x)

r <- imgRedBand(x)
plot(r)
image(x[,,1])

g <- imgGreenBand(x)
plot(g)
image(x[,,2])

b <- imgBlueBand(x)
plot(b)
image(x[,,3])

Visual example:

redPal <- colorRampPalette(c("black", "red"))
greenPal <- colorRampPalette(c("black", "green"))
bluePal <- colorRampPalette(c("black", "blue"))

x11(width=9, height=2.5)
par(mfcol=c(1,3))
image(x=seq(ncol(r)), y=seq(nrow(r)), z=t(r), asp=1, xaxt="n", yaxt="n", bty="n", xlab="", ylab="", main="red channel", col=redPal(256))
image(x=seq(ncol(g)), y=seq(nrow(g)), z=t(g), asp=1, xaxt="n", yaxt="n", bty="n", xlab="", ylab="", main="green channel", col=greenPal(256))
image(x=seq(ncol(b)), y=seq(nrow(b)), z=t(b), asp=1, xaxt="n", yaxt="n", bty="n", xlab="", ylab="", main="blue channel", col=bluePal(256))

Answer 3

I like the approach via R's biOps package. After loading your data into canvas, you're able to convert your jpg file from imagedata to raster and do some further processing. Here's my code:

# Required packages
library(biOps)
library(raster)

# Load and plot data
data(logo)
jpg <- logo

plot.imagedata(jpg)

# Convert imagedata to raster
rst.blue <- raster(jpg[,,1])
rst.green <- raster(jpg[,,2])
rst.red <- raster(jpg[,,3])

# Plot single raster images and RGB composite
plot(stack(rst.blue, rst.green, rst.red), 
     main = c("Blue band", "Green band", "Red band"))
plotRGB(stack(rst.blue, rst.green, rst.red))