Spatio-Spectral Total Variation Filter

Features:

Removes Gaussian and sparse mixed noise.

Steps:

1.      Open a file and select Filtering and Enhancement → Spatio-Spectral Total Variation (SSTV) Filter.

2.      Select SSTV filter parameters from the pop-up dialogue window.

a.      The value of parameter λ adjusts the de-noising strength corresponding to sparse noise, whereas parameters μ and ν provide the tradeoff between retaining the original image and smoothness by total-variation regularization, respectively.

b.      The default parameter values λ = 0.1, μ = 0.2, and ν = 0.2 have been found empirically (see the ref. below). These parameters can be adjusted to get the desired denoising strength.

c.      The algorithm seems to be weakly sensitive to the specific values of these parameters and allows a broad range of values. The number of iterations = 40 seems to be optimal to achieve convergence.  

3.      After completion, click OK and visualize the filtered image.

4.      Visualize the results. You can also run Image Statistics → Image Quality → Image Quality (BRISQUE). The improvement in the image quality is shown in the chart. A lower level of BRISQUE score indicates a higher quality of the image.

Notes: The algorithm is computationally demanding. Decreasing the number of iterations might significantly decrease the computation time.

Tip: Optimize the filter’s parameters on a smaller size by first cropping the image.

Additional Information:

Gaussian noise is a kind of signal noise that has a probability density function (pdf) equal to that of the normal or Gaussian distribution. Major sources of Gaussian noise in digital images arise during acquisition e.g., sensor noise caused by poor illumination and/or high temperature, electronic circuit noise and others. The sparse noise includes random valued impulse noise, salt-and-pepper noise, and horizontal and vertical deadlines.

References:

H. K. Aggarwal and A. Majumdar, "Hyperspectral Image Denoising Using Spatio-Spectral Total Variation," in IEEE Geoscience and Remote Sensing Letters, vol. 13, no. 3, pp. 442-446, March 2016, doi: 10.1109/LGRS.2016.2518218.