Literature Reviews - Segmentation of Hand Bone for Bone Age Assessment - page 27

Fig. 2.6 General framework
of edge-based segmentation

Discontinuities
Detection

Edge Operators

Meaningful Edges Spurious discontinuities

Linked

Boundaries

pixels which undergo abrupt changes in gray level intensity, known as discontinu-
ities, defined by derivative values that surpass a pre-selected threshold. However,
not all detected discontinuities are meaningful in the context of segmentation [ 53 ].
The meaningful discontinuities are termed as 'edges', which are a set of pixels that
appertain to the boundary of objects. All these edges, subsequently, are connected
and linked to constitute closed boundary; pixels within the closed boundary are then
labeled as objects. This type of labeling is referred to edge-based segmentation [ 33 ].
Therefore, edge-based segmentation usually deals with two major technical
problems [ 54 ]: (1) local edge feature; the definition of discontinuities and mean-
ingful edges, such as the features and threshold adopted to define discontinuities.
(2) Global edge linking; the linking procedure connects detected binary edge pix-
els to form linked edges. In this sub-section, some classic edge detectors would
be described briefly and several classic edge-based segmentation algorithms are
explored (Fig. 2.6 ).

2.4.1 Edge Detectors

Edge refers to pixels within digitalized image with large gradient, or in layman
terms: image points that undergo sharp variations. Edges detection is an operation
that converts image into a set of meaningful curves that exhibit certain characteris-
tics or features. This operation is critical in filtering out the meaningless informa-
tion while retaining the vital objects structural properties.
Edge detectors are kernels used in edge detection operation to measure each
pixel's neighborhood and quantify the discrete differentiation of the edge tran-
sition include the degree of intensity changes and direction of the changes.