Optical EngineeringNoise estimation in spherical coordinates for color image restoration
|Format||Member Price||Non-Member Price|
A new adaptive filter is presented, designed for the restoration of color images corrupted with an arbitrary noise model. The proposed filter is based on order-statistics restoration techniques and performs a linear combination of a number of difference measures to estimate the noise present in each pixel. To reduce color artifacts, the filtering is performed in a color space that decouples the intensity and color information. The choice of which color space is arbitrary; however, from experimentation it has been found that a spherical coordinate representation of the RGB pixel vectors provides the best results. As the results show, the proposed filter outperforms the current state of the art techniques for impulse, additive, and multiplicative noise types. In addition to the proposed filter, a new color performance measure is introduced based on the mean square error; however, it can be calculated using any error measure that employs differences between two input signals.