A 2D+t Feature-preserving Non-local Means Filter for Image Denoising and Improved Detection of Small and Weak Particles

A feature-preserving non-local means (FP-NLM) filter has been developed recently for denoising images containing small and weak particlelike objects. It explores the commonly used non-local means filter to employ two similarity measurements taken in the original greyscale image and a feature image which measures the particle probability in the original image. In this paper, we report a new approach to image mapping for constructing the feature image by incorporating both spatial and temporal (2D+t) characteristics of objects. We present a 2D+t FP-NLM filter based on the improved particle probability image. Experiments show that the new filter can achieve better balance between particle enhancement and background smoothing for images under severe noise contamination and has a greater capability in detecting particles of interest in such environments.

• Type: Journal
• Year: 2010
• Where: BMVC
• Authors: Lei Yang, Richard Parton, Graeme Ball, Zhen Qiu, Alan Greenaway, Ilan Davis, Weiping Lu
• Keywords: BMVC 2010, Computer Vision, Images, Non-local Means, Particle Probability