54
Z. Amini et al.
W i3D
1
4
⎡
⎢
⎢
⎢
⎣
I −I I I
−I I I I
I I I −I
−I −I I −I
⎤
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎢
⎢
⎣
F hgh
F ggg
F ghh
F hhg
⎤
⎥
⎥
⎥
⎥
⎦
.
(3.14)
The complex coefficients can be calculated in 3D as start dictionaries of 3DCWDL by:
F C2D
1
16
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
I −I −I −I
I −I I I
I I −I I
I I I −I
I −I I I
I I I −I
−I I I I
−I −I I −I
I I I −I
I I −I I
I −I I I
I −I −I −I
I I −I I
I −I −I −I
−I −I −I I
−I I −I −I
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
F hhh
F ggh
F ghg
F hgg
i.F hgh
i.F ggg
i.F ghh
i.F hhg
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
(3.15)
3.4.4 Results
Datasets from two different OCT imaging systems are utilized [16] which includes
Topcon 3D OCT-1000, and Cirrus Zeiss Meditec. Each set is consisted of six randomly selected 3D OCTs. Subjects in Topcon dataset are diagnosed to have Retinal
pigment epithelial detachment (PED), and the ones in Zeiss dataset are diagnosed to
have symptomatic exudates associated derangement (SEAD). OCT data for Topcon
dataset is obtained from Feiz eye hospital, Isfahan, Iran; and Zeiss dataset is provided
by Retinal Image Analysis Laboratory of Iowa [45]. The performance of 11 different
speckle reduction methods are compared [16] (described in Table 3.6).
The value of the measurements for each denoising method on 144 randomly
selected slices is summarized in [16] and results for 72 randomly selected slices
acquired from Topcon OCT imaging is provided in Table 3.7. Some samples of
datasets are also depicted after application of denoising methods in Fig. 3.5. As it
can be seen in Table 3.7, the performance of the proposed method in 3D R/I CWDL
are considerably better than other methods in CNR and Equivalent Number of Look
(ENL). Performance of all of the studied methods are similar in EP and 2D SDWT
had the highest Texture preservation (TP) which is obtained because of the least
difference appeared in the resulted image which cannot be considered as a positive
point. TP is low for 3D CCWT which shows unwanted flattening of this method.