Region Growing (2D/3D grayscale)

Recursive region growing algorithm for 2D/3D grayscale images with polygon and binary mask output

Description

A recursive region growing algorithm for 2D and 3D grayscale image sets with polygon and binary mask output. The main purpose of this function lies on clean and highly documented code.

Usage:
[P, J] = regionGrowing(cIM, initPos, thresVal, maxDist, tfMean, tfFillHoles, tfSimplify)

Inputs:
- cIM: 2D/3D grayscale matrix
- initPos: Coordinates for initial seed position
- thresVal: Absolute threshold level to be included
- maxDist: Maximum distance to the initial position in [px]
- tfMean: Updates the initial value to the region mean (slow)
- tfFillHoles: Fills enclosed holes in the binary mask
- tfSimplify: Reduces the number of vertices with line simplification

Outputs:
- P: VxN array (with V number of vertices, N number of dimensions). P is the enclosing polygon for all associated pixel/voxel
- J: Binary mask (with the same size as the input image) indicating 1 (true) for associated pixel/voxel and 0 (false) for being outside

Acknowledgements

Region Growing and Line Simplification inspired this file.

• Required Products: Image Processing Toolbox
• MATLAB release: MATLAB 7.12 (R2011a)
• Other requirements: Optional: Line Simplifcation by Wolfgang Schwanghart

[Isabel]

 

[gang]

thks very much good job!

[Hanbo Chen]

 

[Yang]

 

[Omer Demirkaya]

The following two statements should be replaced.
sumSQR = meanI*meanI; % sum of squares

stdI = sqrt((sumSQR - N*meanI*meanI)/(N-1)); % update standard deviation

[Omer Demirkaya]

Daniel good work.
I have a suggestion about an efficient mean and standard deviation calculations within your while loop.

Here is the inintilization of the parameters outside the loop.

meanI = regVal; % Mean intensity
sumSQR = 0; % sum of squares
stdI = 0; % Standard deviation
N = 1; % number of pixels

And then within your while loop after the statement,

J(xv+i, yv+j, zv+k) = true;

one can add the following lines of code to progessively calculate mean and standard deviation
gI = cIM(xv+i, yv+j, zv+k);
sumSQR = sumSQR + gI*gI;
meanI = (N*meanI + gI)/(N+1); % update mean
N = N+1; % increment number of pixels
stdI = (sumSQR - N*meanI*meanI)/(N-1); % update standard deviation

Omer.

[Daniel]

Hello Michael. Unfortunately, that's true. I think the easiest way is to give over a handle to the desired axis as an additional input argument.

[Michael]

Hey Daniel, running the 2d example with no initial seed produces an error if multiple figures are open, because get(himage,'XData') returns a cell array, and is then used in axes2pix() which accepts only numerical arrays. I'm not sure of the best way to deal with this, aside from including a 'close all' in the code.

[maram]

i want to segment a mammographic image(from mias database. after choosing the seed the result is like a curve!

[Daniel]

Hello moram, could you please provide some further informations (e.g. what exactly are you trying to do)?

[maram]

it works very well with the example associated!!but once i tried on another image i failed!

[Daniel]

The selection of a seeding position currently works only for 2D input images, since the function does not know, in which slice you may set the initial point. One (easy) way is to check whether the input image is 3D and show up the medial (or a random) slice.

[Gomathi C]

when i try this code for a 3d image setting the initpos value as [] while calling, it gets the input seed point from me. But, unfortunately I get the following error.

??? Attempted to access initPos(3); index out of bounds because
numel(initPos)=2.

Kindly help me to figure out what I have to do to make the program run.

[Ramachandran M.G.]

good work

[Ramachandran M.G.]

 

[Daniel]

Hello marwa. I'm sorry, i am not sure what you mean with "the same characteristics". Do you mean within the same intensity-range? You are welcome to contact me via mail. With regards, Daniel

[marwa hadhoud]

this is a good work, but i need to ask you, if there is a possibility to make this code search on all the image about all the separated regions in the image that have the same charactristics.

[Jerod Rasmussen]

This is great! Worked intuitively right out of the box! Could use an academic reference if available.