STATISTICAL NEAREST NEIGHBORS FOR IMAGE DENOISING

Abstract

Non-Local-Means image denoising is based on processing a set of neighbors for a given reference patch. Few Nearest Neighbors (NN) can be used to limit the computational burden of the algorithm. Resorting to a toy problem, we show analytically that sampling neighbors with the NN approach introduces a bias in the denoised patch. We propose a different neighbor’s collection criterion to alleviate this issue, which we name Statistical NN (SNN). Our approach outperforms the traditional one in case of both white and colored noise: fewer SNNs can be used to generate images of superior quality, at a lower computational cost. A detailed investigation of our toy problem explains the differences between NN and SNN from a grounded point of view. The intuition behind SNN is quite general, and it leads to image quality improvement also in the case of bilateral filtering. The Matlab code to replicate the results presented in the paper is freely available.

Existing System

Non-Local-Means (NLM), 3D Block-Matching (BM3D) denoising filter

Proposed System

Here this system proposes an alternative strategy to collect neighbors, named Statistical NN (SNN), which reduces the prediction error of the estimate of the noise-free patch. When filtering real images, SNN tends to blur low-contrast image details with a low signal-to-noise ratio more than NN; then explain this drawback of SNN resorting to our toy problem to analyze the differences between NN and SNN from a statistically grounded point of view, and show that a compromise between the NN and SNN strategies is easily found in practice. Our analysis and experimental results, show that, using fewer neighbors, SNN leads to an improvement in the perceived image quality, as measured by several image quality metrics on a standard image dataset, both in case of white and colored Gaussian noise. In the latter case, visual inspection reveals that NLM with SNN achieves an image quality comparable to the state-of-the-art, at a much lower computational cost. Then finally show that the intuition behind SNN is indeed quite general, and it can be applied to bilateral filtering, also leading to an image quality improvement.

Architecture

BLOCK DIAGRAM