椭圆窗口和参数自适应的非局部均值算法

对于传统的非局部均值(NLM)算法,方形搜索窗口会将大量低相似度的图像块引入去噪图像的加权平均计算过程中,导致去噪图像的细节轮廓变得模糊.针对此问题,提出了利用控制核函数来获取椭圆窗口和图像块参数的自适应NLM算法.首先,根据图像的局部梯度信息和结构张量获得可描述图像局部边缘结构的椭圆方程,并由此确定搜索窗口的形状,从而将搜索窗口的搜索范围限制在与图像局部结构相一致的区域内;然后采用控制核函数获得和搜索窗口形状一致的椭圆形图像块,并结合平滑参数自适应的思想进一步增强算法效果.通过在不同噪声等级的经典灰度图中进行实验,实验结果表明,该算法相比于传统NLM算法和参数自适应的NLM算法,在客观的图像评价指标上,有着更高的PSNR和SSIM值;而在主观视觉上,随着噪声等级的提升,该算法在抑制噪声的同时,能够更好地保留住图像的高频纹理信息.

Non-Local Means with Elliptical Window and Adaptive Parameter

In the classic non-local mean algorithm(NLM),the square search window is fixed to be square,which is highly likely to involve many image patches with low-similarity into the weighted averaging process.This way would blur the final denoised image.To this end,an elliptical search window that is defined along the local edge structure is adopted,as well as the strategy of adaptive patch size and smoothing parameter method to improve the ability of noise suppression and detail preservation.Firstly,according to the local gradient information of the image and the structure tensor,the ellipse equations along the local edge structure of the image are obtained,and the shape of the search window is therefore determined.In this way,the search window is refined to the area consistent with local structure.Then steering kernel is leveraged to obtain an ellipse image patch.Finally,the strategy of smooth parameter adaptation is also combined to further improve the performance.Compared with the traditional NLM algorithm and adaptive NLM algorithm,the experimental results show that the proposed method has higher PSNR and SSIM in the qualitative measurement,with the increase of noise level,the proposed method can suppress the noise and retain the high-frequency texture information of the image.