Literature Reviews - Segmentation of Hand Bone for Bone Age Assessment - page 30

uneven illumination and low contrast, especially on the distal phalanges. Besides,
the edge detectors are sensitive to noise and therefore it is not reliable enough
to dispose spurious edges that are not part of the anatomical boundaries of hand
bone. In addition, the edge linking process heavily relies on the precursor image
supplied from previous stages of edge detector processing. Lastly, the information
within the object boundaries is not fully utilized to produce more promising result
of segmentation. Therefore, in the next sub-section, segmentation that makes use
of region information to group similar pixels with respect to particular attrib-
utes into meaningful regions that represent the object are discussed. This region-
based interpretation is important as coherent regions that invariably correspond to
objects in the image.

2.5 Region-Based

The edge-based segmentations discussed in the previous sub-section attempt to
perform object boundaries extraction in accordance to the identified meaning-
ful edge pixels. Region-based segmentations, on the contrary, seek to segment an
image by classifying image into two sets of pixels: Interior and exterior, based on
the similarity of selected image features. In this sub-section, several classic meth-
ods belong to this category are explored and studied.
The region based segmentation is based on the concept that the object to be
segmented has common image properties and similarities such as homogenous
distribution of pixel intensity, texture and pattern of pixel intensity that is unique
enough to distinguish it from other object [ 33 ]. The ultimate objective is to parti-
tion the image into several regions where each region represents a group of pixels
belong to a particular object.

2.5.1 Seeded Region Growing

One of the famous region methods is seeded region growing; this method grows
from seeds which can be regions or pixels, then the seeds expand to accept other
unallocated pixel as its region member according to some specified membership
function [ 76 ]. The details of both regions growing method are illustrated in subse-
quent paragraphs.
Suppose a few groups of pixels or a few pixels are chosen as seed regions or
seed pixels to expand to a region coherently, denoted by R I . Let M ( x , y ) depicts set
of all unallocated pixels in the image, described by the mathematical expression as
follows [ 77 ]:

n n

M = ( x , y )∈ R i N ( x , y ) ∩ R i = ∅

(2.7)

i = 1 i = 1