ABSTRACT
In this chapter we review recent work in image denoising, deblurring, and super resolution reconstruction based on the nonlocal means approach. The nonlocal means (NLM) filter was originally proposed for denoising by Buades, Coll, and Morel [1, 2]. The basic NLM filter is a form of weighted averaging over a large window, where the weights are determined by the similarity between image patches or neighborhoods. The NLM algorithm has a very simple form, has few parameters, and is easy to implement. Its denoising performance is very good and was considered state-of-the-art at the time of its introduction. The NLM filter is able to remove additive white noise while preserving sharp edges and fine texture details. The approach attracted significant attention in the image processing community and was quickly improved upon using refinements and iterative extensions [3-5]. The state-of-the-art in denoising is currently formed by a group of methods that are related to NLM [6-9] and are discussed later in the chapter. Furthermore, the NLM approach has been applied in other signal processing areas such as video denoising [10], image segmentation [4], super resolution [11], and deblurring [12]. Due to the high computational cost required by the NLM filter, several variations of the algorithm have been proposed that aim to reduce the time complexity [13-15].
