Convert RGB-image to one-channel gray image in R EBImage

Question

An R-beginner question:

I wanted to convert an RGB-Image to grayscale and display/plot it with image()

library(EBImage)
orig = readImage("c:/test/b/s2.png")
gray = orig
colorMode(gray) = Grayscale

display(gray) #works
image(gray) #Error 'z' should be a matrix

The image converted by colorMode(gray) = Grayscale seems to be incompatible with image-function. Does Crayscale image in R EBImage has more than one channel?

Then I converted it manually and was able to call image() on it

r = channel(orig,"r")
g = channel(orig,"g")
b = channel(orig,"b")

gray1 = 0.21*r+0.71*g+0.07*b

display(gray1) 
image(gray1) #works

However, the images both gray differed slightly concerning intensity. Is there a way to convert RGB to one channel gray in R EBImage?

EDIT For answering the question, why EBImage:

The package provides some image-processing functions. E.g. I could display easily the intensity graph (img2) of a scanned test-stripe (img1) using further EBImage commands:

blotgraph = resize(gblur(gray1,3),200,1)
plot(blotgraph, type="l")

I was not aware how to solve such kind of tasks without EBImage

Solution 1

Sorry if I am wrong, but as I understand your conversion to greyscale is not being done in a proper way. In general with EBImage conversion to grayscale is done with the command grayimage<-channel(yourimage,"gray") and image(grayimage) gives

Image
  colormode: Grayscale 
  storage.mode: double 
  dim: X Y 
  nb.total.frames: 1 
  nb.render.frames: 1 

where the colormode gives the nb.total.frames:3

Solution 2

This might be simpler, though there are faster and cleaner ways to compress the color layers.

library(png)
foo<-readPNG("c:/test/b/s2.png")
#that's an array n by m by 3 . Now reduce to grey
bar<- foo[,,1]+foo[,,2]+foo[,,3]
# normalize
bar <- bar/max(bar)
# one of many ways to plot
plot(c(0,1),c(0,1),t='n')
rasterImage(bar, 0,0,1,1)

Note that this has created a greyscale image object bar while 'saving' the full-color object foo .

Comments

[Scott Ritchie]

In the code you've posted you've made a mistake in the first block. Should be image(gray) not image(grey)?

[Carl Witthoft]

Why do you want to use EBImage ? Converting image objects to grayscale is not a big deal with the standard base and graphics package tools.

[Valentin H]

@Manetheran OK, thanks! I've fixed. However it has of cause not solved the problem :-).

[Valentin H]

@CarlWitthoft: I came to R more or less accidentally. I've googled for some image-processing related issues and came across EBImage. I've tried R and loved it instantly. So you would suggest to try implementing everything with core-functions first?

[Valentin H]

Interesting to learn png-package! bar<- foo[,,1]+foo[,,2]+foo[,,3] without factors? If you've just leaved the factors for the sake of example-simplicity, then it is the same as: gray1 = 0.21*channel(orig,"r")+0.71*channel(orig,"g")+0.07*channel(o‌​rig,"b"). I thought, there is an explicit command for rgb to gray 8-bit conversion. Thank you anyway!

[Carl Witthoft]

@ValentinHeinitz there must be an rgb--to--gray function somewhere,but I can't find it just now :-(

[Valentin H]

Perfect! That's what I've been looking for. (I think you've meant nb.total.frames:1 not 3) I was not aware of "gray" parameter-value for channel-function