Temporal non-local-means filtering in hybrid 3D endoscopy

Abstract

Time-of-Flight (ToF) cameras are a novel and fast developing technology for acquiring 3D surfaces. In recent years they have gathered interest from many fields including 3D endoscopy. However, preprossessing of the obtained images is absolutely mandatory due to the low signal-to-noise ratio of current sensors. One possibility to increase image quality is the non-local-means (NLM) filter that utilizes local neighborhoods for denoising. In this paper we present an enhanced NLM filter for hybrid 3D endoscopy. The introduced filter gathers the structural information from an RGB image that shows the same scene as the range image. To cope with camera movements, we incorporate a temporal component by considering a sequence of frames. Evaluated on simulated data, the algorithm showed an improvement in range accuracy of 70% when compared to the unfiltered image.