Image denoising using non-local means for Poisson noise

Abstract

The non-local means method is a high-performance noise reduction method that utilizes the structural similarity of an image. The non-local means method generally assumes the noise is Gaussian, and the noise strength is distributed evenly over an image. In the normal non-local means, the weighting function for the noise reduction strength is controlled by a single fixed parameter. However, the non-local means method is not suitable for application to X-ray images, due to the existence of Poisson noise, in its current form. In this paper, we propose an image denoising method using non-local means for an image with Poisson noise. The weighting function in the proposed method adjusts the weight parameter based on the estimated noise strength from the pixels in a local region. As a result, the proposed method provides good noise reduction performance for Poisson noise without recourse to a variance stabilizing transformation. We demonstrate that the noise reduction of the proposed method is an improvement of 0.1–0.9 dB compared to the standard non-local means.