PERFORMANCE COMPARISON OF MODIFIED HYBRID MEDIAN FILTER WITH LOCAL PRESERVING AND CONVENTIONAL MEDIAN FILTER FOR RANDOM-VALUED IMPULSE NOISE REMOVAL

Abstract

Digital images are prone to corruption during their acquisition, storage, or transmission in noise environment. A specific type of noise which is the random-valued impulse noise (RVIN) corrupts digital images by replacing most the original pixels with two random values within the grey level intensity. RVIN is a variation of fixed-valued impulse noise (FIN) which is similar to RVIN with the exception of the pixel values replacing the original pixels are fixed at grey level 0 (black) and 255 (white). The standard median filter (MED) is commonly used in the removal of impulse noise due to its effectiveness and high computational efficiency but it’s reliability sharply declines at high noise densities. MED also filters each pixel present in an image regardless of whether it is noisy or noise-free, resulting in the image reproduced being blurry and less-sharper than the original image. This paper introduces a new method of RVIN filtering of digital images. This approach is based on two phase of operation which is first the detection phase, followed by the filtering phase. Both phases work in a sequence to remove RVIN from corrupted digital images. The quality of noise removal and image reproduction is assessed by comparing the original image with the filtered image both qualitatively and quantitatively. Results obtained shows that this method able to produce a significantly better results than MED by only targeting the noise pixels. This method can filter the RVIN from the corrupted image while preserving the original image detail quite reliably.