Region Growing
The process of forming a region-based image description (or approximation)
is referred to as segmentation. Image segmentation can be described as the
partitioning of an image into a number of disjoint segments or regions based
on pixel grey level characteristics. There are three main classes of image
segmentation techniques: statistical classification, edge detection, and
region growing. The segmentation to be used here is of region growing
type. A region is defined as an area in the image whose pixels share a
common properties such as similar grey level values. Region growing is the
process of joining neighboring pixels into larger regions based on these
properties.
The region growing used in our segmentation method consists of multi phases, using a more
sophisticated model in each phase. First, all neighboring pixels with
luminance difference less than a threshold are merged. Second,
the mean and variance of each
region is included in the model. A merging cost based on the mean and
variance of each region is computed for each pair of neighboring regions.
All neighboring regions having a merging cost smaller than a specified
threshold are merged in the order of increasing cost.