Image Denoising using Fast Bilateral Filter

Abstract

Image de-noising is the process of mapping a noisy image to a noise free image. The major problem of image denoising algorithms is that edges are vanished after the de-noising process. The bilateral filter is a non linear filter that does spatial averaging without smoothing edges. It consists of a spatial kernel as well as a range kernel to limit averaging to the neighbourhood pixels having similar intensity. The range kernel operates by acting on the pixel intensities. This makes the averaging process non linear and computationally intensive, mainly when the spatial filter is large. Those algorithms whose execution time is independent of size and shape of filter are commonly known as constant time or O(1)algorithm. Here we will see how averaging algorithms can be leveraged for realizing the bilateral filter in constant time, by using trigonometric range kernels.