48
Z. Amini et al.
Fig. 3.3 The block diagram of the proposed method [49]
in subcategories of data adaptive and non data adaptive will be more elaborated by
providing details of sample methods and corresponding results.
3.3 Statistical Model for OCT Contrast Enhancement
3.3.1 Method
In [49] we presented a new statistical model for OCT images and used it for contrast
enhancement of these images. This model is based on a nonlinear Gaussianization
transform and tries to convert the probability distribution function (pdf) of each OCT
intra-retinal layer to a Gaussian distribution.
As mentioned before, OCT images suffer from speckle noise as a multiplicative
noise and using a logarithm operator can approximately convert this noise to additive
Gaussian noise [32, 50]. Hence, OCT data is firstly transformed to the logarithmic
domain and the proposed method which includes three main blocks is implemented
in logarithmic space. Finally, the exponential operator is applied to the enhanced
data. Figure 3.3 shows the block diagram of the proposed method whereas in the first
block after logarithm transform, we try to find a suitable well-fitted mixture model
and its parameters for the OCT image. Because of layered structure of the retina and
also monotonically decaying behavior of the OCT intensities in each layer [10], a
Normal-Laplace mixture model is chosen as a suitable prior distribution for OCT
images. Now each component of this mixture model is gaussianized with specific
mean and variance, so that each enhanced component can be obtained in the second
block. These components are combined with each other in the third block to construct
the entire enhanced image. Indeed, in this block all of the gaussianized components
are combined by the Averaged Maximum A Posterior (AMAP) method.
The pseudo code of the proposed algorithm for contrast enhancement is summarized as follows:
1. Transform image to the logarithmic domain by logarithm operator
2. Fit a Normal-Laplace mixture model to the data using Expectation-Maximization
(EM) algorithm
Z. Amini et al.
Fig. 3.3 The block diagram of the proposed method [49]
in subcategories of data adaptive and non data adaptive will be more elaborated by
providing details of sample methods and corresponding results.
3.3 Statistical Model for OCT Contrast Enhancement
3.3.1 Method
In [49] we presented a new statistical model for OCT images and used it for contrast
enhancement of these images. This model is based on a nonlinear Gaussianization
transform and tries to convert the probability distribution function (pdf) of each OCT
intra-retinal layer to a Gaussian distribution.
As mentioned before, OCT images suffer from speckle noise as a multiplicative
noise and using a logarithm operator can approximately convert this noise to additive
Gaussian noise [32, 50]. Hence, OCT data is firstly transformed to the logarithmic
domain and the proposed method which includes three main blocks is implemented
in logarithmic space. Finally, the exponential operator is applied to the enhanced
data. Figure 3.3 shows the block diagram of the proposed method whereas in the first
block after logarithm transform, we try to find a suitable well-fitted mixture model
and its parameters for the OCT image. Because of layered structure of the retina and
also monotonically decaying behavior of the OCT intensities in each layer [10], a
Normal-Laplace mixture model is chosen as a suitable prior distribution for OCT
images. Now each component of this mixture model is gaussianized with specific
mean and variance, so that each enhanced component can be obtained in the second
block. These components are combined with each other in the third block to construct
the entire enhanced image. Indeed, in this block all of the gaussianized components
are combined by the Averaged Maximum A Posterior (AMAP) method.
The pseudo code of the proposed algorithm for contrast enhancement is summarized as follows:
1. Transform image to the logarithmic domain by logarithm operator
2. Fit a Normal-Laplace mixture model to the data using Expectation-Maximization
(EM) algorithm