Graphics Reference
In-Depth Information
'
'
'
'
[
T
((x , x )),
T
((x , x )),
T
((x , x )),
T
((x , x ))]
∈∈ ,which is a four-bit
binary description, there could be totally 16 kinds of values. Construct a index table, as is
shown
x
ˁ
1'
, x
ˁ
2
11
22
33
44
i
j
i
j
in
[],
find
the
corresponding
position
in
the
index
table
index
=
bin dec vote
2(
_ )
data
+ ,where
1
bin dec
represents a conversion from binary to
2
decimal. Define
, which is a 16-dimensional vector, count all the
ˑ =
(0, 0, 0, 1 ,..., 0)
index
ˁˁ
1
,
2
symmetry point pairs in
to fill in the corresponding position of
ˑ
to construct a
j
j
16-dimensional histogram descriptor, which we call
CS
−
FREAK
(, )
ʡ ʡ
.
12
x
x
vote_data
index
P
x
x
0000
1
0001
2
0010
3
'
x
x
'
4
0011
4
P
x
'
2
x
'
.
.
1
.
.
.
.
1110
15
1111
16
− ʡ ʡ
Fig. 2.
The construction of the CS-FREAK descriptor
CS
FREAK
(, )
12
Eucli-
dean distance correlation between the key point in the training image and the corres-
ponding one in the image to be matched, if the correlation is larger than a threshold,
then take one vote , count votes, assume the point pairs with most votes as the correct
matching result.
The specific steps to conduct the two-step matching are as following:
Step1. Extract multi-scale FAST key points;
Step2.Construct the simplified 5-layer model and build FREAK descriptor;
Step3.Conduct the rough match with the utilization of FREAK descriptor,
1 if
the hamming distance between the key point in the training image and a number of key
points in the image to be matched is proximity, keep these point pairs and step into
Step4;
2 if the distance is large, which indicates the descriptor has strong identifica-
tion force, correct match result is the point pair with minimum hamming distance;
Step4. Calculate the vote data of the candidate matching point pairs obtained in
Step3;
Step5. According to the vote results, the correct matching result is the one with the
most votes.
Based on the first step matching results, calculate the
CS
−
FREAK
(, )
ʡ ʡ
12
Search WWH ::
Custom Search