Content-Based Retrieval of Aurora Images Based on the Hierarchical Representation - Advanced Concepts for Intelligent Vision Systems - page 250

(a) (b) (c) (d)

Fig. 2. Process of obtaining the edge image of an aurora: (a) Input aurora image, (b) Image after
thresholding, (c) Image after median filtering, and (d) Edge image

2.1 Curve Segment Extraction Networks

The segmentation and representation system uses four curve extraction networks [3]
denoted by N 1 , N 2 , N 3 , and N 4 to detect all instances of various curve segments in a
binary edge image. N 1 is capable of extracting the curve segments for which the slope
along the curve is in the range of [-

, -1]. Similarly, N 2 , N 3 , and N 4 detect the curve
segments for which the slopes are in the range of [-1, 0], [0, 1], and [1,

∞

], respec-
tively. The network outputs a data packet for each curve segment detected. The
packet consists of the starting point of the curve segment, the number of pixels along
the curve, and a binary string which encodes the curve segment using 1-bit chain code
(which encodes diagonal direction using 1 and non-diagonal direction using 0).

∞

2.2 Line Detector

The line detector [3] using two state machines M 1 and M 2 partitions a curve segment
produced by the curve extraction network into line segments. The straight line char-
acteristics are used to partition each curve segment into line segments. The first state
machine M 1 receives a binary string for a curve segment from a curve extraction net-
work. It breaks the input binary string into several disjoint parts such that each part
consists of strings of zeros separated by single ones or strings of ones separated by
single zeros depending on the slope of the curve. And, it counts the number of zeros
separated by single ones or the number of ones separated by single zeros. The result-
ing string of counts is called the characteristic sequence of the curve segment. The
second state machine M 2 processes the characteristic sequence produced by M 1 to
partition the curve segment into straight line segments.

2.3 Hierarchical Graph/Tree Representation

The output of the segmentation system is represented by a hierarchical graph/tree data
structure [3]. Each curve is represented by a node in the graph in which an edge indi-
cates that the two corresponding curves are connected. Each node of the graph is the
root node of the tree data structure that represents the corresponding curve. Each
curve segment of the curve extracted by a curve extraction network is represented by
a node on the first level (Level-1 node) of the tree and nodes on the second level
(Level-2 nodes) of the tree identify the line segments into which a curve segment is
divided by the state machines.