Denoising camera data - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 5

Table 1
Results of the Luminance Estimation

Gaussian h L [ 13 ] Virtual ADA LSLCD [ 13 ]

PSNR 36.84 36.74 36.76 36.85 36.66

PSNR of the denoising result using the respective luminance estimation method.

3.2 LPA-ICI for Neighborhood Estimation
Once we obtained a continuous luminance estimation we can apply the LPA-ICI method [ 7 ]
to find the dimension of the local homogenous neighborhood. The LPA-ICI method chooses
a polynomial model (LPA) of a certain scale. Based on the ICI rule, the scale of the model is
chosen and this scale defines the extent of a shape around each pixel, in which no singularities
or discontinuities are present.
The LPA-ICI method is applied in eight directions. In each direction θ k a set of directional
kernels with the varying scale h is used to find an interval D .

(2)

Γ > 0 is a tuning parameter, which adjusts the size of the interval. The standard deviation of
the estimate is calculated by multiplying the standard deviation σ of the input with the

norm of the used kernel:

(3)

The standard deviation of the input, σ , is calculated using Equation (1) with the signal value x
estimated by using the noisy observation, thus the raw data pixel value. This is a very simple
estimate, but we found that the improvement using a beter approximation is marginal.
The largest possible scale h i is chosen using the ICI rule.