3 Speckle Noise Reduction and Enhancement for OCT Images
65
Fig. 3.10 A sample B-scan
and proposed ROIs. MSNR
and CNR reported in
Table 3.9
Noise ROI
SEAD ROI
Intra-layer ROI
3.5.2 Results
The proposed despeckling algorithm in Table 3.8 was applied on 20 3D OCT datasets
in the presence of wet AMD pathology (SEAD) and MSNR and CNR were measured.
The Region of Interest (ROI) region was defined within the SEAD as shown in
Fig. 3.10.
Table 3.9 shows MSNR and CNR of selected ROIs. In addition CNR curves for
156 selected ROIs in Fig. 3.11 show the SNR improvements of various versions of
proposed method. It can be concluded that non-homomorphic BiGaussMixShrinkL
method outperforms other methods.
Another way for evaluation of proposed despeckling algorithm is investigation
of the performance of intralayer segmentation algorithms before and after applying
our despeckling algorithm. Figure 3.12 illustrates this comparison for the segmented
layers of a 650 × 512 × 128 Topcon 3D OCT-1000 imaging system by applying
the proposed method in [74]. It is observed that although the first layer and layers
under inner/outer segment junction cannot be detected before despecking, they are
detectable truly after despeckling by our algorithm.