CAJ | 학술논문

利用噪声小波系数父子和图像小波系数父子在四叉树上的不同传播特性，首先，将小波系数区分为：噪声系数、图像系数和噪声与图像共存的系数(简称共存系数)。然后，将噪声系数置零，图像系数完整保留。最后，利用小波域内子带能量分布的方向聚类性，采用Stein无偏风险估计，为每个子带确定最佳大小的椭圆方向邻域窗，通过最小均方误差准则在该窗内对共存系数进行去噪估计。实验结果表明，该算法实现了信号和噪声的有效分离，提高了真实信号系数方差估计的准确度，在去除噪声的同时尽可能多地保留了图像的边缘细节，提高了恢复图像的PSNR值。
이용조성소파계수부자화도상소파계수부자재사차수상적불동전파특성，수선，장소파계수구분위：조성계수、도상계수화조성여도상공존적계수(간칭공존계수)。연후，장조성계수치령，도상계수완정보류。최후，이용소파역내자대능량분포적방향취류성，채용Stein무편풍험고계，위매개자대학정최가대소적타원방향린역창，통과최소균방오차준칙재해창내대공존계수진행거조고계。실험결과표명，해산법실현료신호화조성적유효분리，제고료진실신호계수방차고계적준학도，재거제조성적동시진가능다지보류료도상적변연세절，제고료회복도상적PSNR치。
Exploiting the different spread characteristics of noise and information coefficients in the quadtree struc-ture, an image denoising algorithm based on elliptic directional windows in wavelet domain is proposed,in which the quadtree structure is first used to divide the wavelet coefficients into“noise” coefficients,“image” coefficients and “ mixture” coefficients. Then those “noise” coefficients are set to zero, and those “image” coefficients are kept completely. Finally utilizing the advantages of multidirection-selectivity, an optimal elliptic directional window is obtained, using the stein unbiased risk estimation in the wavelet domain, and those “mixture” coefficients are estimated in this optimal elliptic directional windows by minimum mean squared error criterion. The experimental results show that this method has effectively separated noise and image details, improved the accuracy of variance estimation, kept more image details and improved the peak signal-to-noise ratio.