Generalized Extreme Value Trimmed Filter for Random Impulse Noise Suppression in Color Images

摘要

Noise suppression is the first prior task in a machine vision and human perception. In this paper, a novel method to remove high volumes of impulse noise is proposed. Our filter has designed based on the concept of a frequency data distribution that is not always symmetric. The filter employs a Generalized Extreme Value (GEV) distribution for fitting pixel values in each sliding window. From GEV approximation, the impulsive noise pixels were trimmed out when their values were more than the minima and maxima threshold values. From the experimental results, the proposed algorithm provided good performance for removing high impulse noise and outperformed when compared with the state-of-the-art methods.