Steve on Image Processing

Digital image processing using MATLAB: digital image representation 4

Posted by Steve Eddins,

Today I'm starting an regular, occasional series with tutorial material on digital image processing using MATLAB. I'm going to look at topics in roughly the order used in the book Digital Image Processing Using MATLAB, Gatesmark Publishing, 2009, by Gonzalez, Woods, and Eddins. (Yes, that's me. It was a lot of nights and weekends.)

Contents

Digital image representation

Let's start with digital image representation. (I do not plan to cover image formation. For that topic, see Chapter 2 of Digital Image Processing, Prentice Hall, 2008, by Gonzalez and Woods.) The common mathematical representation of an image is a function of two continuous spatial coordinates: \( f(x,y) \). The value \( f(x_0,y_0) \) is often called the image intensity at \( (x_0,y_0) \), although that term is used loosely because it often does not represent actual light intensity. The term gray level is also commonly used.

You can represent a color image mathematically in a couple of different ways. One way is to use a collection of image functions, one per color component, such as \( r(x,y) \), \( g(x,y) \), and \( b(x,y) \). Another way is to use a vector-valued function, \( {\mathbf f}(x,y) \).

Many functions in the Image Processing Toolbox also support higher-dimensional images. For example, \( f \) might be a function of three spatial variable, as in \( f(x,y,z) \). Image Processing Toolbox documentation calls this a multidimensional image.

Converting the amplitude values of \( f \) to a set of discrete values is called quantization. Capturing those values at discrete spatial coordinates is called sampling. When \( x \), \( y \), and the values of \( f \) are all finite, discrete quantities, we call f a digital image.

Coordinate conventions

It's important to pay attention to different coordinate system conventions that may be used in image processing. Often the center of the upper-left pixel is considered to be the origin, (0,0). There is more variation in the assignment of \( x \) and \( y \) axes. The coordinate system in the diagram below, with the \( x \)-axis pointing down and the \( y \)-axis pointing to the right, is used in Digital Image Processing Using MATLAB in order to be consistent with the Gonzalez and Woods book, Digital Image Processing.

From Figure 2.1, Digital Image Processing Using MATLAB, 2nd ed. Used with permission.

When displaying images in MATLAB, the usual convention is for the center of the upper-left pixel to be at (1,1), the \( x \)-axis to point to the right, and the \( y \)-axis to point down.

Images as matrices and arrays

Digital images are very conveniently represented as matrices, which happens to be great for working with in MATLAB. A monochrome image matrix looks like this:

   f(1,1)  f(1,2)  ...  f(1,N)
   f(2,1)  f(2,2)  ...  f(2,N)
     .       .            .
     .       .            .
     .       .            .
   f(M,1)  f(M,2)  ...  f(M,N)

(This happy marriage of a convenient digital image representation and the MATLAB strength at working with matrices is, more or less, the reason I ended up working at MathWorks.)

We represent color images in MATLAB as multidimensional arrays. For example, an RGB image (with three color components) is represented as an M-by-N-by-3 array. A CMYK image (with four color components) would use an M-by-N-by-4 array.

Multidimensional images are also represented in MATLAB as multidimensional arrays. Therefore we rely on context to distinguish between an RGB image (M-by-N-by-3) and a three-dimensional image with three z-plane slices (also M-b-N-by-3). For example, if you pass an M-by-N-by-3 array to rgb2gray, it is clear from the context that the input should be interpreted as an RGB color image and not as a three-dimensional image.

For more information

For more information, see Section 2.1 of Digital Image Processing Using MATLAB. See also the section Image Coordinate Systems in the Image Processing Toolbox User's Guide.


Get the MATLAB code

Published with MATLAB® 7.12

4 CommentsOldest to Newest

Hey Steve, Thanks for the post and the pointers. I always had this issue in not having the image (or imagesc) in the “pixel coordinate system” – I’m dealing with vector data in head slices and I would love to be able to have a 1-on-1 correspondence between the value of my Image matrix pixel and the one displayed on the screen by imagesc.

I’m still struggling with this, and plan to write a little function which does the following 2 things:
1) get(gca,’currentpoint’) => and swap the x,y
2) edit “text update function” of datatip

Is there a better work-around converting from defaulted spatial coordinate system (of imagesc) to pixel coordinate system? Please let me know your thoughts, would appreciate it.

Thanks,
- Vick

Hey Steve,I think this articl is an excellent work! Even old materials could be cast in lights.
I’ve worked with IPT about 2 years,but I just know little about 3D reconstruction/ 3d Display.And I cannot understand the contents of “help isosurface”and other 3D Visualization documents because I know nothing about the Algorithms of 3D R/D(such as Marching Cubes,etc). The materials about these algorithms are hard to learn. SO can you,our master, write sth about?

Sharing—I don’t know that much about 3D reconstruction and visualization algorithms myself. But I’m not sure why you would need to understand the underlying algorithms in order to use a function such as isosurface.

These postings are the author's and don't necessarily represent the opinions of MathWorks.