Local Variance Switching Gaussian Filter
Ali Degirmenci I. CANKAYA Ömer Karal Recep DEMIRCI
AbstractFiltering can be used for different purposes in image processing. One of them is to reduce noise in the image. In this study, local variance based on switching filter is designed to remove Gaussian noise from gray scale images. During filtering process, local variances of each pixel is calculated and then pixels are classified as five clusters according to their local variance values by the k-means clustering method. Depending on the result of the clustering, variance of the Gaussian filter kernel is tuned. In the smooth regions, in which variances of the pixels are low, higher standard deviation Gaussian filter is applied. Higher variance pixels represent the edge pixels, therefore lower standard deviation Gaussian kernel is applied to preserve the edges. In clusters with medium variance pixels, the standard deviation value in the Gaussian filter is changed depending on the local variance values. Experimental results show that designed local variance based switching filter gives better performance to remove the Gaussian noise at various levels compared to the classical filters.