Random valued impulse noise removal using adaptive neuro -fuzzy impulse detector

Abstract

In this paper a two stage image filtering scheme is presented. In the firest stage adaptive neuro-fuzzy system(ANFIS) based impulse noise detector is used to locate the noisy pixels and in the second stage improved vector median filter is used to provide the accurate value of the corrupt pixel. The training of the detector is accomplished by several natural images. The filtering stage changes the pixels in the image when found corrupted by the noise. Results from the computer simulations indicate that the proposed scheme performs much better than other variants of vector median filters. For quantitative measurements peak signal to noise ratio (PSNR) is used for error in luminance values and normalized color difference(NCD) is used to measure the error in chrominance values of the image.