How to imread() an RGB image to a [3 rows by n columns] double form?

I confess I'm struggling with array manipulations -- call me a newbie in Matlab.

Here's the gist of my project, nothing fancy, to read an RGB image as data points to plot in 3D using scatter3.

I created a simple 4 x 3 pixel RGB image in Photoshop, just to experiment.

(I had to resize the original image for Matlab Community)

So, I start with :

img=imread('4x3_RGB.tif');

Then I rescaled the 0 to 255 RGB values from 0 to 1.0 with :

img_double=double(img)./255;

Then I convert RGB to Lab using with :

Lab=rgb2lab(img_double);

At this point, I get this 4 x 3 x 3 double array :

Lab(:,:,1) =

53.5850 53.2408 97.1393

87.7347 91.1132 77.9855

96.5375 60.3242 32.2970

89.2650 68.0698 14.7631

Lab(:,:,2) =

0 80.0925 -21.5537

-86.1827 -48.0875 16.7037

0 98.2343 79.1875

-12.5051 -66.1018 48.1573

Lab(:,:,3) =

0 67.2032 94.4780

83.1793 -14.1312 16.5673

0 -60.8249 -107.8602

43.9621 51.3645 -61.6661

Trouble is, this "format" will not work with scatter3. I need to massage the data to get it in this form :

I tried Reshape, Permute and a few others to no avail. I just can't conceptualize this transform. I was about to try my hand at some kind of loop, to "convert" the data into a new [4 rows by 12 columns] 'matrix' but, before I embarked in this adventure, I decided to use Excel to enter the data in an empty sheet, then copy and paste into a new Matlab variable. I know, shame on me... But at least, I was able to get a "test of concept".

I guess I could continue doing this (tedious) approach but I'm curious what am I missing? How can I massage my Lab data into a series of column vectors to feed into scatter3?

Answer 1

[Voss]

You mentioned you made attempts to use reshape and permute, which is the way I would do it. Does this:

Lab_reshape = reshape(permute(Lab,[3 1 2]),3,[]);

seem to produce the expected result?

[Roger Breton]

Wow! This is exactly it!!!!! I never suspected I had to contatenate the two operations!!!!!

Here's the result :

Lab_reshape =

53.5850 87.7347 96.5375 89.2650 53.2408 91.1132 60.3242 68.0698 97.1393 77.9855 32.2970 14.7631

0 -86.1827 0 -12.5051 80.0925 -48.0875 98.2343 -66.1018 -21.5537 16.7037 79.1875 48.1573

0 83.1793 0 43.9621 67.2032 -14.1312 -60.8249 51.3645 94.4780 16.5673 -107.8602 -61.6661

I'm flabbergasted! Thank you SO VERY, VERY MUCH!!!

On to some fun stuff, now...

Next step, in my project, is to bring in the "Gamut" of some "color device" as a wire-mesh.... To show my students what an Out-of-Gamut color physically means.

[Voss]

No problem!

reshape will go along the first dimension first (i.e., down the first column, then down the second column, etc.), so you've got to permute first to get the first dimension to be the RGB dimension, which is how you want the result to be. Then reshape to "collapse" the other two dimensions into one. It's worth pointing out that the following command:

Lab_reshape = reshape(permute(Lab,[3 2 1]),3,[]);

(note: [3 2 1] vs [3 1 2]) will also give a matrix of the right size with the RGB values in rows, as intended, but the pixels will be in a different order (corresponding to going across the first row of the image first, then 2nd row, etc., rather than down the 1st column, 2nd column, etc. as the first command does). It won't matter for scatter3, but for other purposes it might, so I thought I should point it out so you can make sure you're using the right command for the purpose you need.

[Roger Breton]

My! That is so considerate of you! I will keep your remark in mind...