The positive and random impulse noise reduction using ann and Gaussian recursive filter

Abstract

In this paper, a new method of filtering is introduced which can be adopted to neural networks, which can be applied to improve the corrupted images by salt-pepper noises. In the first step, neural networks are used to identify the location of the noises in image and in the next step; the identified noisy pixel will be reduced by using Gaussian recursive filter. This method is called the Neural Network Gaussian (NNG) filter. Using neural networks to recognize the location of salt-pepper noises prevents incorrect recognition of noise and increase the quality of noise reduction process. Moreover, by using Gaussian recursive filters against the typical median filter, which used only to omit trivial noises, the algorithm performance will also be improved significantly.