• The ORB algorithm [31] is like SIFT divided in two major parts: the detection
and the creation of a descriptor. ORB is based on the Features from Accelerated
Segment Test (FAST) algorithm for the interest points detection and on the
BRIEF algorithm [32] for the descriptor creation. The FAST algorithm relies on
the corner detection method.
The BRIEF algorithm itself is based on the DAISY algorithm [33] which uses
another type of descriptor which is a binary vector which created by following 3
steps:
– The research of a pattern around the interest point,
– The selection of several couple of points a
– The creation of the descriptor itself starting by a comparison of point intensity in
each couple. If the value of the first point intensity is higher of the second, the
value returned is 1, else it is 0. This way allows to obtain a binary chain whose
the size is depending of the number of couples selected.
In the case of the binary descriptor, the pattern used can be different according to
the algorithm.
• BRISK [34] is based on FAST algorithm for detection and inspired of
BRIEF/DAISY for the descriptor creation, it used a concentric pattern to
determined the neighborhood of the interest point.
• The Descriptor-Nets or D-net [35] is based on SIFT for detection, but guided
random position are also proposed. The descriptor creation relies on paths
connecting interest points in graph. Thus, it uses the information between nodes
to create the descriptor instead of using neighborhood of interest points.
7.3.2.3 Comparison of Interest Points Between Two or More Image:
Match and Alignment
Once the points of interest of two images are detected, the next step is to establish
the match between the points of the first image with those of the second image.
This match between is determined by computing the Euclidian distance. The
couple of interest points with the smallest distance is preserved. The difference
between the two Euclidian distances of the interest points descriptor selected is then
computed. The Fig. 7.15 is a diagram of the possible matches between three
descriptors of three interest points.
The difference obtained is next compared with a user defined threshold and the
match is considered strong when the difference computed is higher than the
threshold. In this case, the point of interest of the first image and the best of
the points selected in the second image are considered the same. This method of
matching is considered as brute-force mechanism.
7 Alignment of Tilt Series
201
and the creation of a descriptor. ORB is based on the Features from Accelerated
Segment Test (FAST) algorithm for the interest points detection and on the
BRIEF algorithm [32] for the descriptor creation. The FAST algorithm relies on
the corner detection method.
The BRIEF algorithm itself is based on the DAISY algorithm [33] which uses
another type of descriptor which is a binary vector which created by following 3
steps:
– The research of a pattern around the interest point,
– The selection of several couple of points a
– The creation of the descriptor itself starting by a comparison of point intensity in
each couple. If the value of the first point intensity is higher of the second, the
value returned is 1, else it is 0. This way allows to obtain a binary chain whose
the size is depending of the number of couples selected.
In the case of the binary descriptor, the pattern used can be different according to
the algorithm.
• BRISK [34] is based on FAST algorithm for detection and inspired of
BRIEF/DAISY for the descriptor creation, it used a concentric pattern to
determined the neighborhood of the interest point.
• The Descriptor-Nets or D-net [35] is based on SIFT for detection, but guided
random position are also proposed. The descriptor creation relies on paths
connecting interest points in graph. Thus, it uses the information between nodes
to create the descriptor instead of using neighborhood of interest points.
7.3.2.3 Comparison of Interest Points Between Two or More Image:
Match and Alignment
Once the points of interest of two images are detected, the next step is to establish
the match between the points of the first image with those of the second image.
This match between is determined by computing the Euclidian distance. The
couple of interest points with the smallest distance is preserved. The difference
between the two Euclidian distances of the interest points descriptor selected is then
computed. The Fig. 7.15 is a diagram of the possible matches between three
descriptors of three interest points.
The difference obtained is next compared with a user defined threshold and the
match is considered strong when the difference computed is higher than the
threshold. In this case, the point of interest of the first image and the best of
the points selected in the second image are considered the same. This method of
matching is considered as brute-force mechanism.
7 Alignment of Tilt Series
201