Improved Non-local Means Using Structural Similarity for Image Denoising

Non-local means (NLM) method is one of the most notable methods in the field of image processing. At first, it is proposed for image denoising. The denoised pixel is obtained by the weighted average of neighboring pixels. Usually, the weight is computed by using the mean square error (MSE). This weight can be called MSE-Weight. In recent years, structural similarity (SSIM) has also been used to design weights in NLM. Unlike them, this paper proposes a novel approach for computing weights by integrating SSIM into MSE-Weight. The advantage of SSIM and MSE is fully utilized. In the two stages of pre-filtering and denoising in the proposed method, the weight of the center pixel in the neighborhood (CW) is carefully selected and designed. The weight in denoising stage is called the MSE and SSIM weight (MSE-SSIM-Weight), and flexibly combines MSE and SSIM. Test results show that the proposed method achieves an average increase of 0.80 dB in peak signal-to-noise ratio (PSNR) and 0.0239 in SSIM compared with the closely related Max-NLM (NLM that takes the maximum weight of neighboring pixels as CW). The comparison with other methods also demonstrates the superior performance of the proposed method.