A GPU-Accelerated Real-Time NLMeans Algorithm for Denoising Color Video Sequences - Advanced Concepts for Intelligent Vision Systems - page 45

A GPU-Accelerated Real-Time NLMeans
Algorithm for Denoising Color Video Sequences

Bart Goossens, Hiêp Luong, Jan Aelterman, Aleksandra Pižurica,
and Wilfried Philips

Ghent University, TELIN-IPI-IBBT
St.-Pietersnieuwstraat 41, 9000 Ghent, Belgium

Abstract. The NLMeans filter, originally proposed by Buades et al.,
is a very popular filter for the removal of white Gaussian noise, due to
its simplicity and excellent performance. The strength of this filter lies
in exploiting the repetitive character of structures in images. However,
to fully take advantage of the repetitivity a computationally extensive
search for similar candidate blocks is indispensable. In previous work,
we presented a number of algorithmic acceleration techniques for the
NLMeans filter for still grayscale images. In this paper, we go one step
further and incorporate both temporal information and color information
into the NLMeans algorithm, in order to restore video sequences. Start-
ing from our algorithmic acceleration techniques, we investigate how the
NLMeans algorithm can be easily mapped onto recent parallel comput-
ing architectures. In particular, we consider the graphical processing unit
(GPU), which is available on most recent computers. Our developments
lead to a high-quality denoising filter that can process DVD-resolution
video sequences in real-time on a mid-range GPU.

1 Introduction

Noise in digital video sequences generally originates from the analogue circuitry
(e.g. camera sensors and amplifiers) in video cameras. The noise is mostly visible
in bad lighting conditions and using short camera sensor exposure times. Also,
video sequences transmitted over analogue channels or stored on magnetic tapes,
are often subject to a substantial amount of noise. In the light of the large scale
digitization of analogue video material, noise suppression becomes desirable,
both to enhance video quality and compression performance.
In the past decades, several denoising methods have been proposed for noise
removal, for still images (e.g. [1, 2, 11, 3, 4, 5, 6, 11]) or particularly for video
sequences (see [7,8,9,10,11,12,13,14]). Roughly speaking, these video denoising
methods can be categorized into:

1. Spatially and temporally local methods (e.g. [8, 11]): these methods only ex-
ploit image correlations in local spatial and temporal windows of fixed size

B. Goossens and A. Pižurica are postdoctoral researchers of the Fund for Scientific
Research in Flanders (FWO), Belgium.