A rule-based instantaneous denoising method for impulsive noise removal in range images

Abstract

To improve comprehensive performance of denoising range images, A rule-based instantaneous denoising method for impulsive noise removal (RID-INR) is proposed in this paper. Based on silhouette features analysis for two typical impulsive noise (IN), dropouts and outliers, a few new coefficients are defined to describe their exclusive features. Founded on several discriminant criteria, the principles of dropout IN detection and outlier IN detection are detailed demonstrated. Subsequently, IN denoising is performed by an Index Distance Weighted Mean filter after a nearest non-IN neighbors searching process. Originated from a theoretical model of invader occlusion, variable window technique is presented for enhancing adaptability of our method, accompanying with practical criteria of adaptive variable window size determination. A complete algorithm has been implemented as embedded modules in two self-developed software. A series of experiments on real range images of single scan line are carried out with comprehensive evaluations in terms of computational complexity, time expenditure and denoising quality. It is indicated that the proposed method can not only detect the impulsive noises with high accuracy, but also denoise them with outstanding efficiency, quality, and adaptability. The proposed method is inherently invariant to translation and rotation transformations, since all the coefficients are established based on distances between the points or their ratio. Therefore, RID-INR is qualified for industrial applications with stringent requirements due to its practicality.