Non-local means image denoising with an adaptive directional spatial filter

摘要

System, apparatus, method, and computer readable media for edge-enhanced non-local means (NLM) image denoising. In embodiments, edge detail is preserved in filtered image data by weighting of the noisy input target pixel value with other pixel values based on self-similarity and further informed by a data-driven directional spatial filter. Embodiments herein may denoise regions of an image lacking edge characteristics with a more uniform spatial filter than those having edge characteristics. In embodiments, directionality of a spatial filter function is modulated based on an edge metric to increase the weighting of pixel values along an edge when there is a greater probability the edge passes through the target pixel. In further embodiments, the adaptive spatial filter is elliptical and oriented relative to a spatial gradient direction with non-uniform filter widths that are based on the edge metric.

主权项

1. An image denoising apparatus, comprising one or more processors to:
receive input pixel values of an image frame; compute an edge metric associated with a target input pixel; compute one or more parameters, based on the edge metric, that specify directionality of a spatial filter function; determine a filtered target pixel value by applying a similarity filter function and the spatial filter function having the specified directionality to the input pixel values; and output the filtered target pixel value to an electronic memory.