Sie sind auf Seite 1von 2

Segmentation Method Overview

Arunesh Mittal

Region Growing
Image segmentation methods are extremely useful in medical imaging as they allow engineers
and clinicians to extract useful diagnostic information from medical images. There are many
classes of image segmentation algorithms and region-growing algorithms are a subclass of
segmentation algorithms. Region-growing algorithms segment an image region into sub regions
based on certain predefined criteria. More formally, the entire set of pixels within an image R is
segmented into n mutually exclusive subsets of pixels denoted by sub-regions Ri s.t. !!!! ! =
.
In region growing, typically, seed regions or sets of pixels are selected for each Ri, either
manually, randomly or based on some predefined criteria. The region-growing algorithm then
grows these initial pixel sets or regions Ri by iteratively evaluating a distance metric between
each region Ri and pixels neighboring that particular region Ri i.e. each pixel neighboring the
region Ri is evaluated for similarity to the pixel set within the region Ri. If the similarity distance
measurement between the region Ri and a pixel neighboring the region is less than a predefined
threshold, then the neighboring pixel is added to the region and the region progressively grows.
This is done iteratively either for a predetermined number of iterations or until all the pixels
within the image have been classified into different regions. Furthermore, due to the nature of
region growing, these sub-regions are topologically connected spaces.
The attached code for this paper demonstrates a simple region-growing algorithm. Let us assume
we have the following image.

We first pick two seed regions to segment the light object from the dark background. Once these
objects are picked, a simple heuristic is used find the threshold values for deciding which
neighboring pixels to add to each sub region. For simplicity the heuristic picked was the
following if the pixel intensity of a pixel neighboring a seed region Ra was closer to the value
of that seed region, relative to any of the other seed regions, then the neighboring pixel was
added to the seed region Ra. Based on this heuristic, starting from the seed point, all the pixels in
the 4-connected neighborhood were evaluated to determine whether they belonged to a sub-

Arunesh Mittal

1 of 2


region. The 4-connected sub-region graph was constructed and traversed recursively using the
above-described heuristic.

The above figures display the progressive growth of the sub regions from the initial seed points.
Reference:
[1] W. K. Pratt, Digital Image Processing 4th Edition, John Wiley & Sons, Inc., Los Altos,
California, 2007

Arunesh Mittal

2 of 2

Das könnte Ihnen auch gefallen