Non-Local Means Denoising Algorithm Based on the Kernelized L2-Norm

Xiaoyan Liu, Xiang-Chu Feng, Xiaoping Li, Zhi-Qin Zhao

Abstract: The non-local means (NLM) has been becoming a prevalent method in image denoising. However, the denoised performance of this method depends heavily on the accuracy of similarity measure. In this paper, we present kernelized L <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -norm to measure the similarity of image patches. In the proposed measure, both the differences of corresponding pixel between image patches and the difference between the center pixel and other pixels in the same image patch are taken into consideration. Compared with the Gaussian weighted L <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -norm, the new similarity measure can protect the image details effectively which leads to the NLM algorithm based on kernelized L <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -norm (K-NLM) generate better denoising results. The experimental results illustrate the proposed method is effective.