Impact of neighborhood size on median filter based color filter array interpolation

Abstract

Color filter array interpolation algorithm play a very important role in obtaining good quality color images from the modern consumer end digital still cameras with single imaging sensors. This paper presents a detailed analysis of impact of neighborhood size on median filter based color filter array interpolation algorithm. The detailed experiments are carried out on all the images of the Kodak image dataset and the image quality is compared using often used quantitative criterion: peak signal-to-noise ratio (PSNR). Results presented here show that generally superior quality images can
be obtained by using 55 neighborhoods in median filtering stage in comparison to traditionally used 33 neighborhoods; around 5.6% improvement was observed on PSNR values on average across all the images considered and all three color channels for each image. Using 77 or 99 neighborhood
in median filtering stage did not improve PSNR values for the most of the images in comparison to using 55 neighborhood