Why is my "grayscale image" a 3D array; can I convert to 2D grayscale without sacrificing quality?

I am working on a reconstruction algorithm that uses total generalized variation (TGV) to reconstruct a partial parallel MR image. I need to use an image that has nice smoothness details (a gradual intensity gradient) in some areas, separated by edges, like that in figure 5 of

http://math.uni-graz.at/mobis/publications/SFB-Report-2009-038.pdf

In fact I am trying to extract the image in 5(a) by taking a screen shot, copying to Paint and converting to .png (or .bmp, or .jpg) to be read in Matlab. The problem is that after I use

I=imread('extracted_image.png');

I is of size n x n x 3 . I figured that MATLAB simply assumes it is a rgb image. I have tried typical ways to get the image to n x n grayscale: I1=rgb2gray(I); I2=sum(double(I),3)./3;

But this induces some streaks across the image. It's really important to keep the smoothness of the textures. I'm not really sentimental to this image but now it's bothering me that I can't get it to work.

Thanks!