Optimized Parameters for Bell-Shaped Error Function in Image Denoising

Year 2017, Volume: 18 Issue: 5, 988 - 999, 31.12.2017

Kemal Özkan , Erol Seke

https://doi.org/10.18038/aubtda.346122

Abstract

Adaptive image denoising algorithms rely on an error
function that measure the distance between an estimated result and
expectations. Selection of the error function and its parameters are crucial
for a successful denoising implementation. In this paper, a method for
determining close-to-optimal parameters for a bell-shaped error function is
evaluated. The function with calculated parameters is employed within a
gradient optimization algorithm and tested using test images with varying noise
types and levels. The restoration results of the denoising test runs that use
the proposed parameters are compared against the results of algorithms that
employ well-known least squares and sum of absolute differences methods along
with a method that combines both. The clear superiority of the bell-shaped
error function for the proposed parameters is shown by the test results.

Keywords

Adaptive iterative restoration, Bell-shaped error function, Data fidelity, geometric tight framelet, Geometric tight framelet

References

• Banham MR, Katsaggelos AK. Digital image restoration. IEEE Signal Proc Mag 1997; 14 (2): 24-41.